ChatGPT

Visa has linked its payment infrastructure to ChatGPT, enabling AI agents to recommend retail products and execute financial transactions. The deployment removes human intervention from the final stages of the retail funnel. Autonomous agents will now process user prompts, evaluate merchant catalogues, and complete the checkout process using Visa’s payment rails at any supporting merchant. Previous retail AI integrations restricted automated purchasing to single-vendor environments. Retailers built proprietary chatbots confined entirely to their own inventory. Visa’s integration bypasses closed-loop architecture. The payment giant connects the open-web reasoning capabilities of a large language model directly to a universal transaction network. Users simply command the agent to procure an item, and the model handles the vendor selection, product comparison, and financial settlement. Enterprises should be aware that commercial transactions will increasingly execute without a human buyer ever seeing a retailer’s website, digital advertisement, or promotional email. Restructuring retail data for AI agent buyers Marketing departments design campaigns around human psychology, emotional triggers, and visual merchandising. AI agents operate on pure data evaluation. When ChatGPT receives a mandate to purchase a specific product type, it parses technical specifications, aggregated sentiment scores, and pricing structures. Display ads and user interface optimisations hold zero weight in the model’s selection criteria. Retailers will need to expose machine-readable inventory data. Search engine optimisation transitions into language model optimisation. The algorithms driving ChatGPT rely on structured data feeds, clear API documentation, and explicitly-formatted product attributes to evaluate whether an item meets the user’s parameters. Merchants failing to maintain high-quality, structured metadata will find their products invisible to the autonomous agent. Personalisation occurs entirely on the user’s device or within the user’s secure LLM profile. The AI retains the consumer’s past preferences, sizing requirements, budget constraints, and brand affinities. Instead of the retailer attempting to guess the consumer’s needs through tracking cookies and site behaviour, the agent arrives at the digital storefront with a highly-specific procurement mandate. Completing a transaction without human intervention requires a secure, automated handshake between the reasoning engine and the payment gateway. Visa provides the financial layer necessary to establish trust in an inherently untrusted agentic environment. Traditional checkout flows require manual data entry, CAPTCHA verification, and two-factor authentication loops. These mechanisms block autonomous agents. Visa implements programmatic tokenisation to solve the authentication problem. The user pre-authorises the ChatGPT environment with specific spending parameters. When the LLM decides on a purchase, it generates a single-use payment token through the Visa network. The agent transmits this token via API to the merchant’s backend systems. The transaction settles exactly like a standard digital wallet payment, bypassing the visual user interface completely. A digital storefront requiring multi-page navigation or mandatory account creation introduces failure points for the agent. Enterprises actively deploying headless commerce architectures possess an advantage. They can process the agent’s payload, confirm stock levels, and execute the payment token in milliseconds. Enterprises track bounce rates, session durations, and cart abandonment to understand consumer behaviour. An AI agent does not browse—it queries an endpoint, extracts the necessary data, and either executes the payment or terminates the connection. Retailers must develop new telemetry to measure agent interactions. Tracking the frequency of API queries from known LLM IP addresses replaces tracking unique human visitors. Understanding why an agent selected a competitor’s product will require analysing the structural differences in product data feeds rather than running A/B tests on website layouts. Customer retention strategies also need adjustment. An autonomous agent evaluates the market fresh with every prompt unless explicitly instructed by the user to reorder a specific brand. Loyalty programmes must be engineered into the payment token or the user’s LLM profile. If the AI cannot automatically apply a loyalty discount during its background calculation, the merchant loses the pricing advantage intended to secure the repeat purchase. Managing and securing the agentic AI supply chain Prompt injection attacks could theoretically manipulate an agent into purchasing from malicious vendors or authorising inflated transactions. Visa’s network acts as the final validation layer, applying fraud detection models to the incoming token requests. Businesses face the secondary challenge of managing automated returns and customer service queries initiated by the AI. If the delivered product fails to meet the parameters defined in the original prompt, the user can instruct the agent to reverse the transaction. In this scenario, the AI will autonomously navigate the merchant’s return policy, initiate the refund request, and generate the necessary shipping labels. Retail customer service operations must deploy their own automated systems capable of negotiating directly with the consumer’s agent. Visa’s ChatGPT integration confirms the enterprise transition from human-operated software interfaces to autonomous digital proxies. The customer is no longer necessarily a human navigating a web browser, but an algorithm executing a script. See also: Aviva deploys AI to stop £230M in sophisticated insurance fraud Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Visa ChatGPT integration enables AI agent retail purchasing appeared first on AI News. View the full article

If your remit is to help your organisation add AI agents to accelerate its processes, you have to start at the foundation – and that means making your data available for AI consumption. Agentic AI scales on data strength, as Niels Zeilemaker, global CTO at Xebia, explains. “If you don’t think about that, you can build the best agent, but it will never be able to find the correct data; maybe it will misinterpret the data, maybe it will join different fields together in your data which should never be connected,” explains Zeilemaker. “And these mistakes are not necessarily the fault of the agent. It’s the fault of your foundation, which is not ready for AI agents.” One area to particularly consider, Zeilemaker notes, is data cataloguing. It’s not a new concept, but the game changes for agents. “If you’re setting up a data catalogue for an organisation only consisting of humans, there’s always a fallback,” he says. “If there’s something not really well documented, you can pick up the phone, walk to a colleague, and have a sort of back door, in ‘how should I work with this particular set of data?’ “Agents don’t have such a back door. They have to rely on the data catalogue, what’s written there, and if the description is wrong, the agents will not perform.” Xebia’s focus is to help organisations turn AI strategy into production-ready solutions which drive real transformation faster. The company’s core values include being people first and quality without compromise, but perhaps the most important, as Zeilemaker sees it, is sharing knowledge – such as at events like TechEx Global North America, at which Xebia is participating. “I think sharing knowledge is very important for us, and it also allows us to be a bit ahead of the curve, adopt quickly to new changes in the market, because everybody has this eagerness to find out new things, and to share what works, what doesn’t work,” says Zeilemaker. “By pushing a lot into this sharing knowledge and innovation, we try to also pick a couple of domains where we want to be the authority.” Data and AI is evidently one of those areas. At AI & Big Data Expo, Zeilemaker will be telling attendees how to build this AI foundation and unify their fragmented data landscapes. The promise is an honest account of how combining purpose-build AI agents with expert engineering compresses a 12- to 24-month timeline into a fixed-price, milestone-bound engagement. The overarching thread for this is what Xebia calls Agentic Data Foundation (ADF), which extends the data platform to host agents, and then make use of them both in customer-facing use cases and internal processes. While there has always been a big appetite in migrating from legacy to modern platforms, Xebia is seeing more customers asking for an approach to more quickly – and reliably – migrate into data platforms. Zeilemaker says this is where consultant and customer are co-developing the solution. “After doing migrations the old-fashioned way, and accelerating some with LLM coding, we are now integrating this into the data platform, making use of the additional context it can provide to accelerate migrations even further,” he says. That accumulated experience is what shaped Xebia Axis: Agentic Data Foundation, Xebia’s answer to helping enterprises make their data AI-ready faster than any alternative. Another weapon Xebia has in its arsenal is Xebia ACE: AI-Native Software Engineering, a framework which embeds AI across an organisation’s entire software development lifecycle (SDLC). Done right, delivery can be accelerated by up to 40%, while legacy transformation costs are cut by up to 70%. Zeilemaker notes that Xebia ACE is particularly useful for larger enterprises who ‘maybe still want to stick to a particular governance or way of working while doing SDLC’. Yet there is a ******* picture here. Zeilemaker uses vibe coding as an example. “If you think about vibe coding, everybody can create an app, but nobody is daring to actually push these apps into production,” he says. “If you adopt ACE, you still get a lot of the benefits of the acceleration of LLMs, but you’re still having the same quality end results as you’re used to in the past. “If you’re looking to make the switch to using LLMs in coding, Xebia ACE will give you a very nice framework to use, without the risk, or any drawbacks of doing dark factory LLM and hoping for the best – and losing a bit of control or governance in the process,” adds Zeilemaker. For enterprises, that control is key. With so much code being generated, the AI-driven SDLC could become a security weakness through vulnerabilities. Zeilemaker argues it’s something the industry still needs to figure out to a degree, but notes with interest the recent move by Anthropic to release a pull request reviewer. “It’s an interesting one, which we’ll probably see more of,” he says. “There will be very lengthy pull request reviews, which you apply whenever you go and try to do a new production release. And then you add a very senior team member in the form of an LLM to your process, which does a sort of third-party review. “I think that’s an interesting angle with what we’re going to see more of in the future.” Ultimately, wherever organisations are in their journey, from assessing their data readiness to being ready to build, Xebia is able to help get the foundations right – and create the transformations on top of it. Photo by fabio on Unsplash Want to learn more about AI and big data from industry leaders? Check out taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Xebia: On building the data foundation for AI agents – and then accelerating appeared first on AI News. View the full article

“We’ve all had that moment where you search for something you know is there, but it just won’t show up.” Apple’s Stacey Ford, vice president of OS Program Management, was talking about Spotlight at WWDC 2026, but she could have been describing the company’s AI ambitions. On Monday at Apple Park, the thing that wouldn’t show up finally did: Siri AI, the assistant rebuilt from scratch after years of underdelivery. The new Siri sustains genuine multi-turn conversation, draws on what’s in a user’s mail, messages and photo library, fields live queries from the web, and carries out tasks across applications. Apple is giving the assistant its own dedicated app alongside system-wide integration, with iPhones showing Siri activity in the Dynamic Island as requests run. That is the version Apple presented on stage. The version worth examining sits in the footnotes: who is actually powering Siri AI, and who gets to use it. Google under the hood Apple’s most consequential disclosure was a quiet one. The company said it collaborated with Google and the Gemini family of models to develop the next generation of Apple Foundation Models that power its Apple Intelligence experiences, the architecture on which Siri AI runs. After two years of insisting its in-house models would close the gap, Apple has answered the question of how it caught up: it didn’t, alone. The company spent considerable keynote time pre-empting the obvious objection. “We believe privacy in AI is non-negotiable,” senior vice president Craig Federighi said, adding that “data is only used to execute your request, and outside experts can continue to verify this promise at any time.” The privacy architecture may well hold. The strategic picture is harder to soften. Apple now depends on its largest search rival for the intelligence layer of its own assistant; at the same time, Google is shipping Gemini across Android, Workspace and its own hardware. Whatever the terms of the arrangement, Apple has conceded that the frontier model race is one it could not win on its own timeline, and that admission carries weight far beyond Cupertino. If the world’s most valuable hardware company, with its silicon advantage and effectively unlimited budget, chose to license rather than build, the sovereign AI ambitions being drafted in capitals around the world deserve a more honest read of what “building our own model” actually costs. The Siri AI rollout map tells its own story Then there is the question of who gets Siri AI at all. The initial beta, due later this year, supports English only. China is off the map entirely, with Apple citing unresolved regulatory requirements, and EU users won’t see the assistant on iPhone or iPad at launch. Apple has said a path forward is being worked on; in the meantime, its updated press release confirms EU availability is limited to macOS 27 and visionOS 27 at first. Read that map from Asia, and the gaps are glaring. China, Apple’s most contested market, is excluded outright, while domestic assistants from ******** vendors ship without restriction. An English-only beta leaves Mandarin, Japanese, Korean, Bahasa and Hindi speakers, which is to say most iPhone users in the world’s fastest-growing smartphone markets, on the old Siri for an unspecified *******. Apple gave no timeline for additional languages. The company that built its reputation on shipping the same product to everyone, everywhere, on the same day, has shipped its most important software in years to English speakers only, minus China entirely and minus iPhone users in the EU.” Catching up, by Apple’s own staging The keynote’s structure was telling. TechCrunch noted that Apple opened by repairing what was broken before showing off what was new, and positioned the upgraded Siri as one entry on a lengthy list rather than the headline act. It was also a transition moment. This was Tim Cook’s final WWDC as CEO before John Ternus, Apple’s senior vice president of hardware engineering, takes over on September 1. “I truly believe the best is still ahead at Apple,” Cook said in his closing remarks. Perhaps. Siri AI is a real product at last, and the demos suggest Apple’s integration instincts remain intact. But Ternus inherits an assistant that thinks with Google’s models and a rollout plan that asks most of the planet to wait. The catching up, it turns out, has only just started. (Photo by Apple) See also: Apple plans big Siri update with help from Google AI Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Siri AI arrives with Google inside, and much of the world is locked out appeared first on AI News. View the full article

McDonald’s is testing a new AI system that can take drive-thru orders and support restaurant operations. The system, called ArchIQ and nicknamed “Archy,” was introduced during the company’s Worldwide convention, according to Restaurant Business. It is being tested at five McDonald’s locations in the United States, though the company has not named the restaurants involved. A video shared on X by a McDonald’s franchise owner showed the system greeting customers, processing order changes, displaying the final total, and asking customers to pull ahead for pickup. A demonstration shared on X by the franchisee account McFranchisee showed the system taking orders in English and Spanish. The account said the system has processed more than one million transactions, with about 90% of orders completed without being escalated to staff. The same account said ArchIQ can respond when repeat customers ask for their usual order. McDonald’s has not provided technical details on how that feature works. ArchIQ is being developed with Google. According to McFranchisee, McDonald’s restaurants in the US are receiving Google Edge Cloud blades ahead of the rollout. McDonald’s previous AI ordering test ArchIQ is McDonald’s latest AI test for drive-thru ordering. The company previously worked with IBM on an automated ordering system across more than 100 restaurants. McDonald’s ended that pilot in 2024 after customer complaints over order errors. The earlier IBM test was followed by customer videos showing incorrect orders, including one case in which the system reportedly added more than $250 worth of chicken nuggets. After ending the IBM partnership, McDonald’s said it would continue exploring voice ordering technology. Restaurant operations support ArchIQ is not limited to customer ordering. McFranchisee said it can monitor restaurants and alert managers to possible issues. According to McFranchisee, the system can alert managers if a freezer is down. It can also flag kitchen bottlenecks or other problems that need attention. McFranchisee described ArchIQ as both an ordering tool and a management-support tool. The test forms part of McDonald’s new growth plan, called “McDonald’s > NEXT.” The company said the plan is intended to improve restaurant operations and unit economics. McDonald’s reported a large digital customer base in its 2025 results. The company said systemwide sales to loyalty members across 70 markets rose 20% to nearly US$37 billion in 2025, while 90-day active loyalty users rose 19% to nearly 210 million at year-end. McDonald’s CEO Chris Kempczinski said in a press release that the strategy is aimed at the company’s next phase of growth and productivity. The company has also referenced restaurant upgrades and possible menu changes under the same plan, but has not provided detailed information. Automation and service In a company memo, Kempczinski said more of the customer journey is becoming automated, leaving fewer chances for guests to interact with crew members. He said that it raises the standard for hospitality when customers interact with staff. QSR Magazine’s 2025 Drive-Thru Report, citing Revenue Management Solutions, said drive-thru traffic remained negative month after month and hovered between minus 5% and minus 8% in 2025. Other fast-food chains have also announced AI-powered drive-thru ordering systems, including Taco Bell and Wendy’s. Jonathan Maze, editor-in-chief of Restaurant Business, told ABC News that companies often present drive-thru automation as a way to free employees for other tasks. The McFranchisee account said the system could reduce the need for workers to take orders in noisy drive-thru lanes. Some X users responding to the ArchIQ demonstration said they preferred interacting with human workers. Others supported a more automated ordering process. McDonald’s has not said when ArchIQ could be expanded beyond the five test locations. The company has said the system is intended to improve speed and accuracy while supporting customers and crew. The company’s AI drive-thru system remains in limited testing. (Photo by Boshoku) See also: Walmart’s AI workflows meet the realities of the balance sheet Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post McDonald’s tests Google-backed AI drive-thru ordering system appeared first on AI News. View the full article

Signing PDFs has become an important task for businesses and individuals alike. Whether you’re handling contracts, legal agreements, or forms, the ability to quickly and securely sign PDFs online is essential. Fortunately, with the rise of online PDF signers, signing PDFs has never been easier. Common challenges in signing PDFs Signing PDFs might seem straightforward, but several challenges often arise during the process. Some of the most common issues include: File compatibility: Not all PDF editors or viewers allow you to easily add a signature, especially if the document is encrypted or password-protected. Document security: Ensuring that your signature is secure and not vulnerable to tampering is critical, particularly in sensitive legal documents. Legal compliance: Making sure your electronic signature is legally valid is crucial, especially for contracts and formal agreements. By understanding these challenges, you can better prepare for a smooth and secure PDF signing experience. Choosing the right PDF signer Selecting the right PDF signer is essential for a hassle-free experience. With many options available, it’s important to choose a tool that meets your needs and ensures your documents are signed efficiently and securely. Key features to look for When evaluating PDF signers, here are some important features to consider: Ease of use: The interface should be intuitive, allowing you to quickly upload documents, sign them, and download the signed copy. Security: Look for a PDF signer that offers encryption and complies with e-signature laws, like the ESIGN Act and UETA. Integration with other tools: A good PDF signer should integrate with cloud storage solutions like Google Drive, Dropbox, or OneDrive, allowing easy access to documents. Multi-signature support: If you need multiple parties to sign a document, choose a signer that supports multi-party signatures. Audit trail: Ensure the signer provides an audit trail for legal purposes, documenting who signed the document and when. Comparing popular PDF signers There are several options to choose from when it comes to PDF signing tools. Here’s a quick comparison of popular PDF signers: Lumin: A comprehensive solution for signing and collaborating on PDFs. Lumin’s easy-to-use interface and robust security features make it a top choice for professionals. DocuSign: A well-known and trusted electronic signature platform with enterprise-level features and extensive legal compliance. Adobe Acrobat Sign: Adobe’s offering integrates with other Adobe tools and provides a reliable, secure way to sign PDFs. HelloSign: Known for its user-friendly interface and ease of use, HelloSign is a great option for small businesses and individuals looking to sign documents online. Each of these tools offers different features, but the best option will depend on your specific needs. Step-by-step guide to signing PDFs online Now that you know how to choose a PDF signer, let’s walk through the process of signing a PDF online, whether you’re using Lumin or another tool. Preparing your document Before you sign your document, make sure it’s ready: Ensure the document is complete: Double-check that the content of the document is final and there are no further edits needed before signing. Check for any required fields: Some PDFs, especially forms, may have fields that need to be filled out before you sign. Ensure document compatibility: Make sure the PDF is not encrypted or password-protected, as this could prevent you from adding a signature. Using a PDF signer tool Here’s a simple guide to signing PDFs online using a PDF signer: Upload the PDF: Open the PDF signer of your choice and upload the PDF document you need to sign. Choose Signature Type: Most PDF signers will allow you to either type your name, draw your signature, or upload an image of your signature. Place the Signature: Drag and drop your signature to the appropriate spot on the document. You may also be able to resize or adjust its placement. Add Initials or Date: If required, you can add your initials or the date of signing. Save and Download: After signing the document, save it and download the signed copy for your records or to share with others. Verifying your signature Once you’ve signed your PDF, it’s important to verify that the signature is correctly applied and that the document is secure. Many PDF signers, including Lumin, automatically ensure that your signature is encrypted and legally binding. Always check the signature’s status before finalizing any legal agreement. Benefits of using an online PDF signer There are numerous advantages to using an online PDF signer, especially when compared to traditional methods like printing and scanning. Time and cost efficiency One of the primary benefits of signing PDFs online is the time saved by eliminating the need to print and scan documents. You can sign documents from anywhere, at any time, which is particularly helpful for remote work and teams spread in multiple locations. And, many online PDF signers offer free versions or affordable subscription plans, which saves you the cost of ink and postage. Enhanced security measures Using an online PDF signer often offers enhanced security features, including encryption and compliance with legal e-signature regulations. Tools like Lumin ensure that the signed document is protected from tampering and provide an audit trail that verifies who signed the document and when. This is especially important for businesses that deal with sensitive contracts or legal agreements. Tips for a seamless PDF signing experience Here are a few tips for a seamless signing experience in order to get the most out of your online PDF signer: Ensuring compatibility Make sure that the PDF signer you choose is compatible with your operating system and the devices you use regularly. Many tools work in multiple platforms, including Windows, macOS, and mobile devices, but it’s always best to confirm before starting. Maintaining signature legality Ensure that the tool you’re using complies with the necessary electronic signature laws, like the ESIGN Act in the US or eIDAS in the EU. These laws ensure that your digital signature is legally binding, so you can confidently sign contracts and agreements online. Final thoughts Signing PDFs online has never been easier, and with the right PDF signer, you can streamline your workflow and ensure that your documents are signed securely and efficiently. Whether you’re using Lumin, DocuSign, or another tool, taking the time to choose the right PDF signer for your needs will help you save time, reduce costs, and improve document security. The post How to sign PDFs easily online with a PDF signer appeared first on AI News. View the full article

Autonomous AI agents are altering the speed at which software is shipped. Unfortunately, they are also shrinking the time it takes for a mistake to become a catastrophe, creating a dangerous blind spot in many security strategies. The threat no longer comes just from external ransomware or malicious insiders. It comes from authorized, internal tools. To make matters worse, these tools cause damage faster, across more systems, and with fewer chances for your security team to notice in time. In 2025 alone, major DevOps platforms experienced 68 distinct AI-related security incidents, ranging from prompt injections to credential exfiltrations. But even more concerning is the trajectory, incidents accelerated significantly in the latter half of the year, as the DevOps Threats Unwrapped 2026 Report shows. Organizations must accept that access controls alone cannot stop an authorized agent from making a destructive mistake. Once an agent is authenticated, access controls assume its actions are intentional, leaving you defenseless if the AI misinterprets a prompt or hallucinates. The pivotal question for your security strategy now is no longer how you control these agents, but how fast your business can recover when they execute a destructive command. The Threat from Within: How AI Data Loss Emerges and Scales Traditional data loss scenarios revolve around predictable adversaries—a developer accidentally deleting a repository or a ransomware group extorting your infrastructure. AI introduces a completely different threat vector. The fundamental problem with AI-driven data loss is that the call is coming from inside the house. This means you must protect your production environment from the tools you explicitly authorized to modify it. Traditional security defenses fall flat against AI-driven data loss for two main reasons: AI agents do not hack their way in; they interact with your environment using the API keys, tokens, and permissions you provide them, executing commands as trusted insiders. An agent can hallucinate, encounter an error, or fall victim to an injected prompt, triggering destructive actions in milliseconds. This isn’t just theoretical. When an autonomous tool goes off the rails with elevated access, the fallout is immediate and severe. In the 2026 PocketOS incident, during a standard workflow, an AI agent tasked with a routine operation stumbled upon a credential mismatch. Instead of halting, it used an unrelated, highly permissive API key left in the environment to erase the production database volume permanently, alongside the provider’s native backups stored in the same blast radius. An entire live production database vanished in exactly nine seconds… This incident proves that when an autonomous agent makes a mistake, the damage outpaces any human ability to detect and intervene, leaving your database exposed to a hyper-accelerated blast radius. And if your recovery strategy relies on human intervention to stop such an agent, it might already be too late. Just as the PocketOS agent had permissive access to database volumes, CI/CD AI agents hold the keys to your version control platforms. If an authorized agent goes rogue, your source code and intellectual property can vanish in seconds, instantly paralyzing development. Ensuring business continuity and operational resilience means fundamentally re-evaluating where your data safety net lives, because your current infrastructure might be a trap. AI Data Loss in DevOps: The Native Infrastructure Trap Assuming that native platform protections will save you from such an AI-driven wipe ignores the fundamental mechanics of the shared responsibility model, where you are responsible for the data. What is more, native platform protection often does not cover deletion and corruption when it is executed by an authorized account. Therefore, relying on your version control platform as your primary backup strategy leaves a massive gap in your disaster recovery plan. Another major engineering flaw seen in DevOps pipelines is the overlapping authorization perimeters. If your backups are stored inside the same platform as your active codebase, they share the same blast radius, as in the PocketOS case. The lesson here is straightforward: You cannot use the same environment to build your code and back it up. Surviving AI-speed threats requires stepping outside the native ecosystem and architecting a truly decoupled backup and DR infrastructure. How to Survive: Architecting a Decoupled Recovery Layer If your native infrastructure is a trap, the only viable survival strategy is physical decoupling. To ensure that machine-speed destruction is met with machine-speed recovery, you must deploy an independent, immutable recovery layer. True resilience against AI data loss requires you to neutralize the AI threat vector across four specific fronts: #1 Blast Radius Isolation AI data loss becomes catastrophic only when an agent’s permissions reach your backups. Physically separate this blast radius by routing your DevOps backups to a completely decoupled storage destination of your choice, such as an independent AWS S3 bucket, Azure, or an on-premise NAS. If an AI agent completely wipes the primary Git environment, the isolated backups remain 100% untouched. #2 Encryption and Immutability An autonomous agent with elevated privileges can easily overwrite business-critical backup storage. Enforcing AES-GCM encryption secures your data against unauthorized access, while WORM (Write Once, Read Many) storage protocols make it systemically impossible for a rogue agent to modify or delete the archive. #3 Complete Context Recovery AI data loss reaches far beyond deletion. It involves subtle corruption, such as when an agent introduces flawed code or poisons a context window. Because source code alone does not restore the full delivery context, you must secure the entire ecosystem, including workflows, pull requests, issues, and pipeline metadata. This allows your team to roll back the entire operational state to a known-good baseline. #4 Granular Restore When AI wipes a repository in nine seconds, time is the deciding factor. Point-in-time granular restore allows DevOps teams to surgically target and recover the exact repositories, branches, or variables the AI agent destroyed, neutralizing the business impact instantly. Securing your source code on these four fronts builds a resilient disaster recovery strategy for your company’s intellectual property. A tested, isolated backup and DR is your secret weapon to maintain business continuity after an AI agent wipes out your repositories. Precaution is Better Than Cure As you integrate more autonomous AI agents into your pipeline, your security strategy must evolve to survive their speed. The only way to act faster than autonomous AI is to act in advance and back up your repositories with a dedicated DevOps backup solution before an AI agent reaches them. GitProtect delivers on all four fronts of AI data loss resilience by enabling you to enforce strict precautionary measures: strict blast radius isolation through BYOS, mathematically unbreakable immutability with AES-GCM encryption and WORM, complete context recovery (both code and metadata), and granular restores. All that secured by robust access controls like RBAC, SSO, and MFA to give you an impenetrable, automated disaster recovery engine. When an agent can erase your environment in seconds, waiting for an alert is no longer a viable strategy. Architectural precaution is the only measure that guarantees your business can recover faster than an AI can destroy it. The post Autonomous AI Data Loss in DevOps: Building Efficient Defenses appeared first on AI News. View the full article

Aviva has uncovered a record £230 million in insurance fraud claims and is using AI tools to counter the growing problem. The battleground has changed, and the culprits are also coming armed with a new generation of tools. We’re now in an environment where AI is being used not just to defend against fraud, but to perpetrate it. The insurance industry has long dealt with opportunistic dishonesty. A bumped car suddenly needs four new doors, or a minor slip becomes a life-altering injury. However, according to Aviva’s data, the nature of the deception is getting deeper, more sophisticated, and harder for the human eye to catch. Aviva is fighting fire with fire, deploying its own AI to uncover these elaborate schemes. Countering the AI-powered insurance fraud factories Aviva reports that scammers are now using AI to generate convincing fakes of car accident scenes. These aren’t clumsy photoshop jobs; they’re detailed, plausible images that can easily fool a claims handler working through a heavy caseload. The same generative AI tools are being used to create fake documents, from invoices for repairs that were never done, to medical reports that have no basis in fact. Fraudsters don’t need access to a network of corrupt garages or medical professionals to back up their story. They just need a subscription to an AI service and a bit of imagination. The AI handles the rest, producing official-looking documents that can pass a cursory inspection. An individual or small group can now generate the supporting evidence for dozens of high-value claims without ever leaving their desk. How do you validate reality when reality itself can be so easily and cheaply faked? Aviva’s response has been to build an AI-powered defence system that can operate at the same scale and speed as the threat. While the company is understandably tight-lipped about the exact architecture, you can piece together what a system like this needs to do. At its core, the AI detective carries out pattern recognition at scale. The AI sifts through millions of data points from current and past claims, learning what a legitimate claim looks like—and, more importantly, what it doesn’t. When a new claim comes in, the system is cross-referencing everything. Does the damage in the photo match the physics of the described accident? Do the timestamps on the documents make sense? Has this vehicle registration number appeared in other suspicious claims? Are the repair costs quoted on the invoice out of line with the thousands of other similar repairs in the database? It’s a level of forensic analysis that would be impossible to perform manually on every one of the thousands of claims filed each day. From organised crime to exaggerated claims It’s important to note that this isn’t all about organised criminal gangs. A portion of that £230 million figure comes from what the industry calls “claims inflation.” Claims inflation is the more common fraud where policyholders or service providers pad the bill. For instance, a garage might add unnecessary repairs to a quote, or an individual might exaggerate the value of items stolen in a burglary. Here, too, AI is proving to be a heavy-duty tool. By analysing vast datasets of repair costs and market values, the system can instantly flag when a quoted price is an outlier. It can compare the cost of a replacement part from one garage against the average from hundreds of others in the same region for the same make and model. The goal of Aviva’s AI isn’t to outright deny claims, it’s an augmentation tool for their human investigators. The AI acts as a filter, sifting through the noise to surface the most likely instances of fraud. This human-in-the-loop approach is essential for ensuring fairness and preventing the system from becoming a ****** box that makes decisions without oversight. What Aviva is doing provides a potential route for any customer-facing enterprise in the age of generative AI. The same technology that creates these threats is also the most effective way to combat them. As it becomes easier to fake everything from identities to invoices, the only viable defence is an intelligent system that can learn, adapt, and spot deception at a scale that humans alone can’t match. See also: Weis Markets adds Instacart AI-powered shopping carts to stores Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Aviva deploys AI to stop £230M in sophisticated insurance fraud appeared first on AI News. View the full article

Weis Markets is adding Instacart’s AI-powered shopping carts, ****** Carts, to select stores in Pennsylvania, bringing digital coupons, loyalty features, and repeat-purchase recommendations into the grocery aisle. The Pennsylvania-based grocery chain is working with Instacart to deploy the smart carts, which include cameras, certified scales, location systems, and a touchscreen. According to Instacart, ****** Carts use basket-facing camera sensors, outward-facing cameras, certified scales, and location-tracking systems to support item recognition and checkout functions. The system combines edge computing on the carts with cloud AI trained on more than 1.6 billion online grocery orders. Shoppers can use the cart screen to monitor spending during their trip. They can also access location-based digital coupons directly from the cart. Weis customers can sign up for a Weis Rewards account through the cart and redeem loyalty benefits while shopping. Customers who link their accounts can also use a Buy It Again feature, which shows items they have previously purchased. Weis and Instacart already work together on online grocery services. In 2023, Weis partnered with Instacart to offer same-day delivery from 133 locations in Pennsylvania, New York, and Delaware. Instacart expands ****** Cart rollout The Weis rollout adds to Instacart’s wider ****** Cart deployment. The company says the carts now span more than 100 cities across 15 states. ****** Carts are available across more than a dozen retail banners, including Kroger, Schnucks, and Wakefern banners such as ShopRite and Fairway Market. Earlier deployments have produced some store-level usage data. Retail Dive reported that Schnucks data showed ****** Carts handled more than 10% of sales on busy days at one store. That store had 10 ****** Carts and around 160 traditional carts, according to the report. Greg Zeh, senior vice president and chief information officer at Weis Markets, described the carts as part of the company’s effort to improve the shopping process. He pointed to real-time spend tracking and on-cart coupons as key features. Instacart described the partnership as an extension of Weis Markets’ use of digital tools inside stores. David McIntosh, Instacart’s chief connected stores officer, said ****** Carts bring together in-store and online data. Weis adds AI to checkout operations Weis has also been adding AI to self-checkout. Toshiba Global Commerce Solutions said Weis completed a chainwide deployment of its ELERA Security Suite across self-checkout lanes. The system includes produce recognition and loss prevention tools. Toshiba says the technology uses edge AI for on-device processing. At the time of Toshiba’s December 2025 announcement, the system was operational across self-checkout lanes in all 199 Weis locations. Weis also reported that more than 94% of customers selected the produce recognition feature at self-checkout. Grocers test AI beyond checkout Albertsons Companies has also introduced an AI-based quality control tool for produce inspection. The system is designed to help identify moldy or damaged fruit before it reaches store shelves. The tool initially focuses on strawberries and red and green grapes. Albertsons says it is intended to improve quality rating consistency and support faster decision-making. The company also says the tool expands quality data and helps align inspections with company standards. Albertsons operates more than 2,000 stores, including Safeway, Jewel-Osco, and ACME. The system supports quality inspectors working in its distribution centres. The quality control system uses computer vision to support produce inspections across Albertsons’ store brands. It was developed in-house by the company’s technology and supply chain teams. Albertsons built the tool on Google Cloud’s Gemini Enterprise platform, including Vision AI and Gemini models. Google Cloud said it advised on the AI component used in the supply chain process. (Photo by Franki Chamaki) See also: Amazon brings AI shopping assistant to retailers with Kate Spade Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Weis Markets adds Instacart AI-powered shopping carts to stores appeared first on AI News. View the full article

Shell will use agents from C3 AI to shift from basic anomaly detection towards fully-automated predictive maintenance. The global energy giant is building on their current use of the C3 AI Reliability Suite, which already keeps tabs on more than 30,000 crucial pieces of equipment across upstream and downstream operations. Shell now intends to lean heavily into autonomous AI agents, putting them in charge of the entire maintenance lifecycle. Going from that first warning sign all the way to a completed repair, this level of automation strips away the need for constant human oversight and makes sure the company’s resources are pointed exactly where they are needed most. “This expanded partnership with Shell proves what’s possible when enterprise AI is fully operationalised at global scale for predictive maintenance—reducing unplanned downtime and delivering hundreds of millions of dollars in economic value,” said Stephen Ehikian, President of C3 AI. “Shell has built mature AI predictive maintenance programs on our platform, and together we’re now pushing into agentic AI, advancing how this technology can further transform reliability, safety, efficiency, and operational performance.” C3’s AI agents help Shell move past basic anomaly detection In the beginning, Shell used machine learning simply to spot odd patterns in sensor data, giving engineers an early heads-up before things broke. To pull this off, the system ingests a massive amount of real-time operational technology (OT) data and mixes it with business context from ERP platforms such as SAP. The next step introduces AI agents built for actual reasoning and independent action. While older systems stopped at pinging an engineer when things looked unusual, this next-generation framework independently investigates why an alert fired in the first place. Once it pinpoints the root cause, the agent steps up to draft precise work orders, confirm part availability in the inventory, and generate procurement requests. C3 AI’s platform handles the heavy lifting, providing a model-driven space to easily integrate high-frequency sensor feeds with structured financial and maintenance logs. These AI capabilities are trained to learn the normal operating baselines for specific gear, like pumps, turbines, and compressors. The agentic layer sits on top of this foundation. Operators configure an individual agent for a given piece of equipment by defining its objectives and permitted responses. If the core machine learning models detect a deviation from normal operations, this agent activates, gathering extensive contextual data to build a complete picture of the situation. This context usually includes recent maintenance history, environmental conditions, and upstream process variables. Using all that information, it suggests a fix backed by solid evidence. Human operators can then easily approve or override the plan. As the system proves itself over time, Shell can fully automate its responses to certain types of alerts. Connecting straight into systems like SAP is critical here, allowing the agent to work inside the exact same workflows that human planners already use. The real impact of agentic AI for predictive maintenance Putting agentic AI to work at this scale tackles the classic “last mile” headache in predictive maintenance. Many industrial companies can predict failures just fine, but turning those insights into fast, efficient action remains a challenge. Usually, engineers still have to manually dig through alerts, investigate the causes, and write up the work orders themselves. Shell wants to shrink that timeline. By letting AI handle root cause analysis and work orders, the delay between a predicted failure and the actual fix drops. That directly improves equipment uptime and protects production. Moving to a model where repairs only happen when the equipment condition actually demands it naturally saves money, simply because nobody is wasting time tinkering with perfectly fine machinery. Leaving healthy hardware alone also means it lasts much longer. On top of the cost savings, stepping in before a catastrophe hits makes the whole operation much safer and cuts down on environmental risks, which is always top of mind in the energy sector. “What Shell and C3 AI have built on Azure over the past several years is exactly what enterprise AI should look like—real applications, running in production, delivering measurable value at global scale,” commented Sandy Gupta, VP GISV, Software Development Companies at Microsoft. This expanded rollout shows that we are finally talking about practical industrial AI production workflows instead of just algorithms. Rather than just the prediction itself, the real value comes from the system’s ability to act on it with barely any human oversight. See also: Meta Business Agent drives AI-powered conversational commerce Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post How C3 AI agents will automate predictive maintenance for Shell appeared first on AI News. View the full article

Meta has launched Business Agent to automate conversational commerce workflows directly inside its messaging applications. The software allows global retail brands to execute transactions and field support tickets without human intervention. Deploying this architecture places agentic AI directly at the core of social commerce. Meta integrated these workflows natively into Instagram, Messenger, and soon WhatsApp. High volumes of customer interactions overwhelm traditional contact centres. Meta’s platform creates a persistent digital sales representative capable of operating globally. The software operates far outside basic chatbot parameters and can execute concrete administrative tasks. How Meta Business Agent collapses the checkout funnel Consumers frequently discover merchandise on Instagram and initiate a Messenger chat regarding sizing variations. The agent intercepts the query and guides the buyer through the checkout process inside the host application. This architectural model eliminates the high cart-abandonment rates associated with external payment portals. Support operations gain massive efficiency by letting the automated system handle repetitive tier-one tickets. Human support staff gain the bandwidth to manage complex account issues. Contact centre directors can reallocate human capital to specialised retention units. Meta markets this capability as an “infinite team” for retail operators. The software assumes full responsibility for initial contact management. It functions as a first-tier response mechanism operating around the clock. Integrating direct business information allows the system to generate highly specific product recommendations. The underlying models learn and adapt from ongoing consumer interactions. Continuous learning improves performance over time without requiring constant manual reprogramming by internal developers. Retailers with seasonal catalogue changes and volatile consumer demands require such adaptability. Product database updates push directly to the conversational interface via automated syncing protocols. Platform-native architecture design Embedding an agent directly within the Meta ecosystem represents a distinct departure from deploying third-party customer service platforms. A native application integrates deeply with a user’s social graph and historical interactions. External API calls struggle to replicate this level of deep consumer profiling. Tight system integration enables secure, in-chat payment processing. Replicating this complex transaction workflow natively remains exceptionally difficult for external vendors. Lower technical barriers accelerate deployment timelines for small and medium-sized operators. However, large enterprises will need to evaluate how this managed service aligns with their existing CRM databases. Software fed with incomplete or poorly structured information generates subpar consumer interactions. Bad automated outputs actively damage consumer trust and corporate equity. Operations teams will need to ensure that support documentation and product details remain clean and machine-readable. Massive corporate data hygiene projects precede any successful product launch. Engineering teams must establish definitive escalation paths. Business leaders determine the exact scope of tasks the automated system is permitted to handle. Hard-coding operational limits prevents unauthorised internal actions. Creating precise handover protocols for human intervention helps to prevent major service outages. Customers trapped in automated conversational loops experience intense brand frustration. Quality assurance teams consume large portions of the pre-launch phase testing these specific escalation triggers. Engineers run thousands of simulated conversations to locate operational edge cases. Security design presents another major implementation consideration. Firms need highly secure authentication methods to verify a customer’s identity before processing returns or checking order statuses. Identity verification adds a heavy layer of process design to the core engineering timeline. Authentication workflows must integrate perfectly with existing internal Single Sign-On providers. Evaluating vendor dependency The core decision for marketing leaders pits adopting a powerful, integrated platform against maintaining an open, custom-built architecture. Selecting the Meta product secures immense distribution advantages. Platform adoption offers a lower initial development cost compared to building architecture from scratch. The target consumer base already exists natively on the application and Meta manages the heavy core processing infrastructure internally. Independent engineering stacks demand heavy internal maintenance and high operational expenditures. However, they offer greater flexibility and long-term application portability. Engineering departments can select distinct large language models for different departmental tasks. Legal teams can dictate exact data residency policies based on regional government regulations. Many organisations will likely deploy hybrid architectural designs to capture the best of both worlds. In this model, platform-native agents serve as a high-volume concierge, handling initial product discovery and routine catalogue routing. Meanwhile, high-value financial transactions and complex account resolutions are seamlessly handed off to proprietary, secure internal systems. By striking this architectural balance, enterprises can capitalise on Meta’s distribution while maintaining the technical autonomy required for long-term operational security. See also: Amazon brings AI shopping assistant to retailers with Kate Spade Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Meta Business Agent drives AI-powered conversational commerce appeared first on AI News. View the full article

Microsoft has announced the wider testing of its new Autopilot feature at the Microsoft Build event this week, backed by a post on the company’s’ website. Autopilots are described as a new category of agents that can work autonomously on a user’s behalf. Microsoft says each Autopilot has its own identity, and so multiple agents can co-exist within different rule sets, letting users run Autopilots at home, or at work, with separate governance and stipulations limiting or allowing specific activities, according to context. Microsoft’s first Autopilot is Scout, which some internal users at Microsoft have been able to test in beta. The project is now being rolled out to “a select group of customers…and Frontier organizations,” according to the company’s blog. Scout’s initial home will be in acting agentically in Microsoft 365 applications, working across Outlook, OneDrive, SharePoint, and Teams, and be able to coordinate data from each platform to schedule meetings, flag important messages, and generate calendar events to keep workers on track with their tasks. Over time, Microsoft Scout learns about each user’s preferences and work patterns, aligning its activities and priorities to become more efficient and tailored. Under the hood, Scout is built using OpenClaw, the vibe-coded project created over the course of a weekend by Peter Steinberger. Microsoft says Scout comes with enterprise-grade security and controls “so it can be trusted in your organization from day one.” Microsoft has stated that it intends to contribute upstream to the open-source OpenClaw project. Administrators whose organisations adopt Microsoft Scout will be able to validate that any Scout implementations operate securely within the bounds of IT and security policies, and be able to validate agent identities via dedicated Entra entries. The agentic platform will be “managed with the same rigor you expect from any first-party Microsoft service,” the company statement reads. The algorithm takes its data protection policy from Microsoft Purview, and the credentials behind a machine identity are redacted from logs and diagnostics to preserve anonymity. Humans are required to sign off on actions deemed sensitive by the algorithm. The early internal trials at Microsoft have allowed it to expose risks to testers using Scout on the desktop, and the company has tuned the agent to balance any security issues found with an ability to “keep work moving without constant prompting.” Letting Autopilots take the burden of low-level tasks can “keep work in motion so it continues even when your attention is elsewhere.” One feature will be to identify deadlines, block book a user’s calendar so preventing other activities from taking place in the run-up to a deadline, and provide the materials necessary to get around what it’s identified as a bottleneck to progressing an important, focused project. The announcement on the Microsoft website was penned by Omar Shahine, Corporate Vice President of Microsoft Scout, a Redmond lifer whose previous experience includes positions in the Windows Live, OneDrive (previously SkyDrive), and Mac Office divisions at the company. Early adopters keen to try Scout will need to be enrolled in Microsoft’s Frontier programme, have an Intune policy configuration, offer an “opt-in attestation”, and have an active GitHub Copilot licence. (Image source: Pixabay, under licence.) Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Scout from M’Soft is the agentic Autopilot that works across M365 appeared first on AI News. View the full article

Amazon is offering its AI shopping technology to other retailers through a new Agentic Shopping Assistant built on AWS, with Kate Spade among the first brands to use it. The service allows retailers to build AI shopping assistants for their own websites and apps. Amazon said each deployment can be customised to a retailer’s catalogue, customer base, shopping environment, and brand voice. The service is based on technology first developed for Amazon’s own online store. Now, it’s packaging architecture, starter code, and lessons from Alexa for Shopping for use by other retailers. More than 300 million customers used Amazon’s AI shopping assistant last year, according to the company, the assistant generating nearly US$12 billion in incremental sales in the same *******. Amazon said the service lets retailers deploy conversational agents “in weeks,” rather than the years building from scratch might take. The offering includes architecture guidance, starter code, and help from AWS experts and system integrator partners. Kate Spade uses AI for gift shopping Kate Spade, owned by Tapestry, is one of the first brands to use the technology, introducing an AI Gift Concierge that helps shoppers find gift options through a conversational interface. The assistant is focused on gift-buying, with Amazon citing its own data that reveals 53% of shoppers “report stress” during gift purchases. The assistant can recommend gifts based on occasion and customer inputs. Fabio Luzzi, Tapestry’s chief data and analytics officer, told Digital Commerce 360 that the tool came from listening to consumers and identifying what they needed when shopping for gifts. Tapestry tested the assistant for about two and a half months before making it available to consumers, according to Amazon’s announcement. AWS services behind the assistant The system uses Amazon Bedrock, AgentCore, and OpenSearch. Bedrock is the basis for generative AI applications, AgentCore operates AI agents, and OpenSearch is for search and retrieval. Amazon said conversational shopping sessions generate conversion rates 3.5 times higher than traditional keyword-based product searches. The AWS launch follows Amazon’s rollout of Alexa for Shopping in the US in May. That allows users to type shopping-related questions into Amazon’s search bar and receive conversational answers. Under the hood, Alexa for Shopping brings together elements of Rufus and Alexa+ (Rufus was the AI shopping assistant Amazon launched in 2024). (Photo by Shutter Speed) See also: AWS’s legacy will be in AI success Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Amazon brings AI shopping assistant to retailers with Kate Spade appeared first on AI News. View the full article

Standardising grid data through SAP S/4HANA allows E.ON to modernise infrastructure and execute AI deployments. The utility giant manages infrastructure across three distinct domains: energy grids, customer solutions, and energy infrastructure solutions. Maintaining operations across this scope requires continuous capital expenditure on IT hardware and software maintenance. Leadership initially questioned the business case supporting large-scale technology spending. The engineering team proved that persistent financial investment guarantees system stability, affordability, and resilience within a digitised energy network. E.ON prioritises growth, sustainability, and digitalisation as primary corporate objectives. Falling behind in technical capabilities carries long-term financial costs. Infrastructure standardisation drives uptime E.ON executes a cloud ERP migration alongside its SAP S/4HANA implementation. Legacy ERP systems in the utility sector often suffer from extreme customisation. The engineering department rejects fragmented custom builds to avoid this technical debt. Developers integrate established software packages directly into a cohesive architecture. This design methodology guarantees data scalability across the enterprise. The focus on foundational infrastructure delivers highly visible production outcomes. E.ON reports a 77 percent reduction in IT downtime over a five-year *******. Achieving these uptime metrics requires standardising data tables and removing redundant middleware from the technology stack. SAP S/4HANA uses an in-memory database architecture. This design choice accelerates query processing times compared to legacy relational databases. The utility provider leverages this speed to process telemetry data streaming from grid assets in real-time. Fast data processing serves as the prerequisite for deploying any machine learning models against operational data. Technology leaders face intense pressure to match the pace of external software development. E.ON CIO Sebastian Weber notes this pressure creates tension. Consumer software sets expectations for enterprise application deployments. Weber finds consumer AI applications like ChatGPT solve domestic problems effectively, creating internal demands for similar workplace automation. The energy company must close the gap between external software capabilities and internal readiness. Internalising data and cybersecurity operations E.ON treats internal readiness as a primary business objective. The company expanded its internal engineering teams aggressively and hired over 1,000 specialists to bring technical capabilities in-house. The recruitment drive secured more than 500 data experts and 300 cybersecurity professionals. Bringing data engineering in-house allows the utility provider to build proprietary data lakes and audit data governance internally. Retaining internal cybersecurity talent ensures the company maintains strict access controls over the operational technology systems managing the physical energy grid. Engineering now acts as the primary vehicle for achieving commercial targets in the European green energy sector. Of course, managing digital ecosystems at this volume requires strict oversight. The technical team establishes centralised governance structures across all business units. Administrators deploy standardised contracting frameworks and unified IT system management consoles. Having such an administrative architecture in place enforces security standards and cost discipline without restricting feature development. Standardising vendor contracts accelerates software procurement timelines while capping runaway licensing costs. Deprecating isolated innovation hubs Enterprises often isolate experimental technologies in separate business units. E.ON completely abandoned this methodology and deprecated experimental garages and isolated digital labs. Management integrates digital tools directly into active business processes. Keeping innovation teams separated from production environments often prevents applications from surviving the transition to live servers. By forcing developers to build within the core architecture, the engineering department guarantees production viability. “Bringing the system up to speed requires internal readiness,” explained Weber. “It means we must think deeply about investments, prioritisation, and most importantly, people and culture.” Weber expects the operational velocity to remain high, noting the company will not return to previous delivery speeds. New software deployments require precise alignment with business requirements. E.ON enforces a “BizDevOps” operating model. This framework forces developers to build features that generate exact commercial value. Engineers collaborate directly with business analysts during the initial architecture phase. This methodology is paired with targeted employee training. Line workers and managers receive specific instruction on operating newly-deployed tools. This capacity building ensures staff can extract verifiable value from the modernised infrastructure. E.ON is taking a pragmatic approach to AI E.ON manages its AI deployments with deliberate caution and refuses to build proprietary AI platforms from scratch. Instead, leadership prefers to leverage partnerships with established technology vendors. This procurement strategy maintains flexibility across the corporate software portfolio. Engineers explore specific, bounded use cases for machine learning applications. The technical roadmap targets customer service automation, predictive maintenance, and operational optimisation. Applying predictive maintenance algorithms to energy grids prevents catastrophic hardware failures. Sensors detect voltage anomalies and transmit the data back to the central S/4HANA instance. Machine learning models analyse this telemetry to identify wear patterns on physical infrastructure. Maintenance crews receive automated dispatch orders before the equipment actually fails. This active mitigation strategy reduces emergency repair costs and prevents localised power outages. Testing these applications via third-party providers prevents the company from overcommitting capital to unproven frameworks. E.ON embeds these automation features directly into core systems rather than treating them as optional add-ons. The technology serves a customer base of 47 million users. Processing user requests through automated customer service workflows reduces call centre loads and accelerates incident resolution. “In essence, our experience highlights a broader truth about digital transformation,” Weber noted. He explained that pushing new software to production cannot compromise system stability, cybersecurity, or governance frameworks. Without proper alignment with business requirements, advanced technologies fail to deliver value. The modernised architecture provides E.ON with the necessary foundation to scale green energy infrastructure reliably. See also: Walmart’s AI workflows meet the realities of the balance sheet Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post How E.ON uses SAP S/4HANA to modernise the grid with AI appeared first on AI News. View the full article

Walmart has reportedly begun limiting employees’ use of an internal AI assistant called Code Puppy after demands placed on the LLM backing the tool were higher than expected. Employees of Walmart were encouraged to use Code Puppy without any stricture or stipulations as to the limits of use, but Walmart is now assigning employees a fixed number of AI tokens, which limits how much it can be used. Code Puppy was publicised as being able to help with tasks like spreadsheet analysis, creating presentations, and other automatable workplace activities. The change in internal policy is a cost control measure, as LLMs are increasingly transitioning to pay-per-use, rather than the fixed-price, subscription model that gave near-limitless access to AI inference. Walmart has roughly 2.1 million employees, so even modest per-employee queries and task requests can create significant costs. Walmart’s guidance to employees is to use AI where it can create value, and comes with guidance on how workers should choose the right AI tool for any given task. Reporting also says employees have access to other AI platforms paid for by the company. Walmart has expanded AI tool use in the company and provided training for its employees in how to use an AI, encouraging workers to experiment and adopt successful uses. Now the costs of each interaction are being billed directly, it’s among other large enterprises struggling to balance reported improvements in productivity with the cost of achieving the same. At least part of the issue may stem from the methods used to measure productivity in workflows based on AI. Previously, tracking the number and complexity of uses of AI tools as measure of productivity has led to many employees ‘gamifying’ their KPIs – so-called ‘token maxxing’. As recently as April this year, a partner at Sequoia Capital told The Wall Street Journal, “We all should be tokenmaxxing”, an approach that resulted the emergence of AI leaderboards in companies to celebrate those making best use of AI software. Such performative practices at companies will increasingly incur costs relative to the number and complexity of AI tasks, and the model chosen to perform them. Larger models that perform recursive actions (‘thinking models’) use more tokens to process inputs introspectively, leading to higher bills for users. Walmart’s encouragement of workers to choose their model carefully is an attempt to limit spending on expensive, frontier models to achieve relatively trivial tasks, such as spreadsheet analysis and creating presentations. Multi-agentic AI work may also create unexpected costs for employers. When employees instigate iterative loops running on multiple agents in order to create a desired outcome, the real cost of sub-optimal results (and the necessary refining and re-submission of prompts) is now measurable in hard cash. While not all AI providers have changed the entirety of their billing models from fixed subscriptions to per-token, both Anthropic and OpenAI have already moved their higher tier enterprise plans to the new footing. Microsoft’s decision to charge for its GitHub Copilot software development tools as of June 1st is in line with what is rapidly becoming the new financial normality for model providers. Uber recently revealed that it had used up its 2026 budget for AI spend in the first four months of the year; a testament to changes in charging policy affecting end-users. By setting limits on token use on a per-employee basis, Walmart is striving to keep a lid on its ongoing costs, enforce more thoughtful use of AI tools, and enable it to establish the metrics of return on investment in AI. (Image source: Pixabay, under licence.) Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Walmart’s AI workflows meet the realities of the balance sheet appeared first on AI News. View the full article

Phishing protection refers to the category of cybersecurity products that are specifically designed to help companies detect, prevent, and respond to phishing attacks before they develop into full-blown data loss situations. Modern phishing attacks are no longer limited to deceptive emails. They now span the full attack chain, including impersonation, cloned websites, session **********, and real-time credential harvesting. AI-generated phishing attacks are making traditional detection methods less effective, increasing demand for adaptive phishing protection technologies. Given how rampant and complex phishing attacks have become, modern tools focus on the problem from different layers of the attack chain, whether that’s adding in protections directly within the inbox, the browser, or taking actions on the spoofed website itself. What Is Phishing Protection? Phishing protection technologies identify and block malicious attempts to steal sensitive information by misleading employees through deceptive communications. In most cases, the end goal of the attacker is to gain access to an account, steal credentials, and make it out with valuable data. One of the most common phishing tactics is when an adversary impersonates a trusted entity to try and trick the target into willingly handing over credentials, financial information, or even just sending out payments directly. Usually, phishing attempts arrive via email, but modern attacks are branching out and are also targeting employees through email, voice calls, fake websites, and cloned login portals. Why Phishing Protection Matters Phishing remains one of the most common causes of credential theft, business email compromise (BEC), and ransomware incidents. As phishing attacks become more personalized and AI-generated, organizations increasingly rely on dedicated phishing protection platforms to detect threats that traditional email filters miss. Even the most tech-savvy people within a company are falling victim to these more sophisticated, AI-powered attacks. This article will highlight some of the best phishing protection solutions in 2026, which include Proofpoint, Abnormal Security, Memcyco, Barracuda, and IRONSCALES. These platforms help organizations prevent phishing attacks through email filtering, behavioral AI, impersonation detection, browser-level protection, and real-time credential harvesting prevention. Best Phishing Protection Solutions to Evaluate in 2026 PlatformPrimary ApproachBest Suited ForProofpointEmail filtering, threat intelligence, VAP targetingLarge enterprises with high email volumeAbnormal SecurityBehavioral AI for BEC and vendor impersonationOrganizations facing targeted social engineeringMemcycoReal-time phishing site and credential harvesting detectionFinancial services, ecommerce, brand-targeted industriesBarracudaGateway filtering, AI inbox defense, bundled trainingMid-sized to enterprise organizations wanting a single vendorIRONSCALESAdaptive AI, SOC automation, phishing simulationSecurity teams integrating detection with employee training 1. Proofpoint Proofpoint is the most widely deployed anti-phishing email security solution among the Fortune 100 companies. Proofpoint scans over 3 trillion emails every year with its Nexus Threat Graph platform, using AI to examine language, visuals, and URL payloads in emails. Their main differentiator is its VAP (Very Attacked People) model, which identifies who is most at risk of being targeted by a phishing attack. This is based on things like their role, seniority, level of access, history, etc. The platform filters emails, isolates browsers, deploys anti-phishing training, and also comes with automated incident response. Best suited for: Large enterprises with high email volumes that need intelligence-driven phishing protection. 2. Abnormal Security Abnormal Security takes a behavioral approach to detecting phishing emails. The platform establishes a baseline of what normal email communications look like for each individual user, vendor, and partner within an organization’s ecosystem. Any email that deviates from these established baselines gets flagged and quarantined. This approach is especially effective for business email compromise (BEC) and vendor email compromise (VEC). According to the 2026 Attack Landscape Report from Abnormal Security, vendor email compromise accounts for 61% of all BEC attacks. Best suited for: Organizations facing sophisticated phishing attacks, particularly BEC attacks and vendor email compromise attacks. 3. Memcyco Memcyco represents a newer category of phishing protection focused on browser-level interception and credential harvesting prevention. It focuses on what happens outside of the inbox, monitoring activity on the spoofed websites and cloned login portals that hackers use to harvest credentials after someone clicks on a phishing link. The platform detects phishing websites and brand impersonation in real time, often before they show up in the threat databases. Memcyco works directly within the browser at the session level, intervening just at the moment when hackers are about to capture login information. The added kicker is that it can swap out sensitive information that employees enter with decoy credentials. This not only renders them useless, but completely turns the tables and gives companies visibility into the hacker as soon as they attempt to log in with the decoy. Best suited for: Companies that are targeted by brand impersonation, especially in the financial and eCommerce spaces. 4. Barracuda Barracuda Email Protection package includes gateway filtering, AI inbox defense, automated response to security threats, and security awareness training. The platform defends against 13 categories of email threats and can be deployed via API to Microsoft 365. The AI engine learns the organization’s communication patterns and can recognize anomalies that may indicate phishing, spear-phishing, or account take over attacks. Any malicious emails are automatically pulled from all affected inboxes. Over 200,000 organizations use Barracuda for their cybersecurity needs. Best suited for: Mid-sized to enterprise-level companies looking for a single vendor to provide them with cybersecurity training and email protection. 5. IRONSCALES IRONSCALES is a cloud-native phishing protection platform that uses adaptive A, automated incident response, and simulation training to protect organizations from phishing attacks. The platform offers protection to over 17,000 organizations through its API integration with Microsoft 365. The latest Winter 2026 release of IRONSCALES introduces three AI agents. The Red Teaming Agent performs reconnaissance on publicly available data to simulate phishing attacks into the detection model. The Phishing SOC Agent investigates reported phishing attacks automatically. The Phishing Simulation Agent generates personalized training calibrated to what an adversary would actually send. Best suited for: Security teams looking for protection from phishing attacks employee training integrated into a single adaptive platform. Phishing Protection and the Broader Security Strategy First off, no phishing protection tool will ever replace the importance of having good fundamentals in cybersecurity. Solid endpoint security solutions, access controls, and incident response teams will always do the heavy lifting for an organization. Phishing protection tools help fill the gaps that these other tools leave behind. With real-time cloning and multi-channel impersonation threats bypassing the many legacy security systems, this is exactly where dedicated anti-phishing protection tools show their worth. These tools allow organizations to detect and respond to threats that would otherwise slip through their existing security system, whether those threats come in the form of a convincing lure that passed through the email gateway, a cloned login page harvesting credentials in real time, or a deepfake voice call targeting a finance team. How to Choose the Right Phishing Protection Solution This really comes down to your current level of exposure. If the biggest risk is email volume and socially engineered messages getting past filters, platforms like Proofpoint, Barracuda, and Abnormal Security are built for that problem. If the risk sits further down the chain, where users are already clicking through to credential harvesting sites, Memcyco covers that stage. If the priority is tying detection directly into training and SOC workflows, IRONSCALES brings those together. Start with the gap your current stack leaves open and work outward from there. Key Takeaways Modern phishing attacks extend beyond email into browsers and cloned login portals Behavioral AI and session-level protection are becoming increasingly important Different phishing protection tools focus on different stages of the attack chain Organizations should choose phishing protection solutions based on where threats bypass their existing defenses FAQs About Phishing Protection What Does Phishing Protection Do? Phishing protection tools detect and block malicious attempts to gain access to sensitive information through deceptive emails, fake websites, and impersonation campaigns. These tools work at different stages of the attack chain, from filtering emails before they enter the inbox to detecting fake login pages and alerting users in real time. How Do Phishing Attacks Work in 2026? In 2026, attacks use generative AI to create phishing emails that are highly personalized to each victim, deploy highly convincing cloned login portals, and impersonate trusted contacts across multiple channels at the same time. The use of AI removes many of the typical spelling and formatting mistakes that employees may formerly have been trained to spot. Can Phishing Protection Stop AI-Generated Attacks? Behavioral AI and session-level detection are effective because they analyze patterns and intent rather than matching known signatures, which fail against unique AI-generated messages. The best tools adapt continuously, learning what normal communication looks like within an organization and flagging anything that deviates from that baseline. What Is the Difference Between Email Security and Phishing Protection? Email security covers spam, malware, data loss prevention, and compliance. Phishing protection specifically targets credential theft and social engineering, often extending beyond email to web and SMS. Many organizations use both, with email security handling the broad filtering layer and phishing protection focused on the targeted, high-risk attacks that slip through. The post 5 Phishing Protection Solutions Security Teams Should Evaluate in 2026 appeared first on AI News. View the full article

Microsoft’s Majorana 2 quantum chip arrived this week with numbers that are genuinely difficult to contextualise: qubits 1,000 times more reliable than the first generation, a mean qubit lifetime of 20 seconds against an industry norm measured in microseconds, and a revised roadmap targeting a commercially scalable quantum computer by 2029. Behind those numbers is Microsoft Discovery agentic AI, and that platform is arguably the more consequential part of this announcement. To put that in plain terms: most quantum chips today can hold their fragile computational state for a fraction of a second before losing it. Majorana 2 holds it for up to a minute. Microsoft’s own analogy is a phone battery that, instead of dying in a day, lasts nearly three years on a single charge. Majorana 2 was developed with the help of Microsoft Discovery, the company’s agentic AI platform for scientific R&D, which also reached general availability this week. The timing is deliberate. The quantum chip is Microsoft’s proof that the platform works. What Microsoft Discovery agentic AI actually did here The common read on this story is that AI designed the chip. The reality is more specific, and arguably more interesting. The decision to switch the superconducting material from aluminium to lead, which Microsoft says is the single change most responsible for the reliability improvement, came out of years of conventional materials research, not an AI recommendation. What Microsoft Discovery’s agents did was everything around that: managing fabrication workflows, automating measurements that previously took weeks each, breaking down nearly two decades of siloed research data, and surfacing correlations that no single researcher could hold in their head across that volume and variety of information. “As you run AI agents on this data, they’re able to essentially resynthesize and make correlations that we as humans cannot see because no single individual has that much vision across that much data,” said Zulfi Alam, corporate vice president for quantum at Microsoft. That framing matters because it shifts the story from “AI built the chip” to something more accurate: agentic AI compressed the experimental cycle. What would have required extensive trial-and-error to find the right atomic-level recipe for the chip’s crystalline structure could, through AI-driven simulation, be narrowed to a single targeted experiment. “In the new world order, through simulations, you can see where the highly probable target is. And then with that knowledge, you ideally only have to experiment once,” Alam said. The measurement problem, solved One of the more concrete wins the team describes involves qubit measurement; the process of detecting quantum states by determining whether there’s an even or odd number of billions of electrons on a semiconductor wire. When done manually, this takes weeks. Microsoft tried to automate it a few years ago using earlier machine learning and couldn’t. With agentic AI built on Microsoft Discovery, they created a specialised agent that now runs the process automatically and continuously, building three-dimensional maps of qubit conditions at a pace no individual researcher could replicate. “Using agentic AI to automate the measurements was a game changer,” Alam said. The agent handles parallel voltage adjustments across hundreds of parameters simultaneously, something human researchers, thinking linearly and structurally, cannot do. Chetan Nayak, Microsoft technical fellow leading the quantum programme, said the shift has been thoroughgoing: “Agentic AI has permeated almost everything we do, it’s just become kind of a very natural part of our workflow.” Microsoft Discovery goes general The platform that underpinned all of this is now available to enterprise customers. Microsoft Discovery combines specialised AI agents for scientific research, a Discovery Engine for research and reasoning workflows, and enterprise-level security and governance. A free Microsoft Discovery app, usable locally with a GitHub Copilot account, is also in early preview, lowering the barrier for individual researchers who want to run the same kind of agentic workflows. The commercial pitch is clear: the same capability stack that the quantum team used to compress its development timeline is now available to any organisation running intensive R&D. Microsoft has already seen uptake in life sciences, chemicals and materials, energy and manufacturing. Syensqo, for instance, is using it to develop next-generation fluids for semiconductor manufacturing. The 2029 claim, in context Microsoft’s revised quantum timeline deserves a note of editorial distance. The company has moved its target from 2033 to 2029 based on Majorana 2’s progress, which is a significant acceleration, but quantum roadmaps have a history of optimistic compression. The 1,000x reliability figure refers specifically to improvements over Majorana 1’s qubits, not a direct benchmark against competing approaches from IBM or Google, which use fundamentally different architectures. Nayak’s own framing is honest about the incremental nature of this: “Where are we relative to last year? We’re 1,000 times better.” That’s a meaningful year-on-year milestone. Whether it holds at the pace required to reach utility-scale quantum computing by 2029 is the question no one, including Microsoft, can yet answer. See also: *** and Germany plan to commercialise quantum supercomputing Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Microsoft’s Majorana 2 quantum chip is also a case study for agentic AI in R&D appeared first on AI News. View the full article

Anthropic’s IPO filing marks the maturation of generative AI from a research-heavy venture phase into a stabilised enterprise utility. Model developers operating in private markets have prioritised rapid iteration and maximum compute performance over predictable billing cycles. Taking a foundational provider public aligns those engineering goals with standard corporate procurement, introducing structured release schedules and established pricing frameworks that decision-makers require for multi-year planning. William Samengo-Turner, Technology Sector Lead at A&O Shearman, said: “If Anthropic pursues an IPO, the most important question isn’t whether public markets are ready for AI—it’s whether AI is ready for public markets.” The enterprise consumer sits directly at the centre of this maturation. Companies integrating Claude into their proprietary workflows can now plan around how public market structures will formalise Anthropic’s pricing tiers, API rate limits, and enterprise service agreements over the coming years. Establishing a public valuation framework Institutions looking to capitalise on generative machine learning have largely invested in hardware providers and infrastructure layers. This indirect approach allowed companies to build out the necessary compute clusters without taking on the concerns around model hallucination or algorithmic copyright disputes. Samengo-Turner notes that public investors have focused on the surrounding ecosystem: “Investors have been able to buy the ‘picks and shovels’ of the AI *****—with infrastructure, semiconductor, and software businesses benefiting from it. Anthropic would offer one of the first opportunities to invest directly in a company building frontier models at scale.” Pricing that asset class presents immense difficulty. Anthropic and its competitors require continuous, massive capital expenditures to train successive model generations. Converting these capital requirements into a public structure introduces high operational drag for both the provider and the client. A public Anthropic will need to balance the need to buy tens of thousands of GPUs against the need to post favourable quarterly earnings, which requires passing those compute costs onto the end user in a predictable manner. Karthik Hariharan, Senior Engineering Manager at DoorDash, commented: “Both OpenAI and Anthropic are racing to IPO ahead of each other and catch up to SpaceX/xAI. The problem is whoever lands first probably sets the floor and ceiling for public market pricing that others will follow for at least 12–18 months.” If Wall Street demands aggressive margin expansion following the IPO, enterprises should anticipate tighter licensing terms and the potential deprecation of older and less profitable model versions. This creates forced migration cycles for corporate development teams, requiring them to constantly update their API integrations to maintain access to the most cost-effective models. The B2B dependency The commercial structure of these public listings relies heavily on enterprise adoption because the consumer market lacks the scale to offset computing costs. Suvrankar Datta, Principal Investigator at ****** Lab, explained: “There are eight billion human beings on the planet… of the eight billion, only 100 million can afford to pay for Claude at the current rate. Even if they pay $20 per month for Claude, it still won’t be able to survive without an IPO.” The $20 monthly consumer tier cannot fund billion-dollar server clusters. Therefore, model providers must extract their required revenue from corporate budgets, integrating their tools into daily enterprise operations such as human resources, legal document review, and customer support triage. Nate Elliott, AI Analyst at Emarketer, said: “We’re about to find out whether the market thinks AI is a consumer story or an enterprise story. Because while Claude has built a solid enterprise user base, it’s just not competitive as a consumer AI platform.” Emarketer forecasts that only 5.4 percent of US internet users will use Claude in 2026, far behind the 36.6 percent who will use ChatGPT and the 27.4 percent who will use Gemini. “The good news for Anthropic: more than 60 percent of US AI users say they use these tools for work, and we believe that percentage will only grow,” adds Elliott. Anthropic will need reliable, high-volume enterprise contracts to demonstrate steady revenue growth to prospective shareholders. Boardrooms can use this dependency to negotiate longer-term price locks and favourable data governance agreements before the public market forces Anthropic to prioritise short-term yield over market penetration. Margin pressures and market consolidation The impending public offering acts as a forcing function for commercial discipline across the entire generative computing sector. Rather than viewing this negatively, enterprises can see it as the end of unpredictable startup behaviour and the beginning of reliable vendor management. Smitarani Tripathy, Social Media Analyst at GlobalData, said: “Discussions reveal increasing concerns around the economics of the AI ecosystem, with several influencers questioning whether massive investments in model development and compute infrastructure can ultimately translate into sustainable profits.” Tripathy further explains that this filing initiates an “AI capital markets race,” where model providers must demonstrate revenue growth, operational efficiency, and defensible business models alongside innovation. If a vendor goes public and fails to achieve sustainable profits, they may aggressively alter their service-level agreements or sunset key API endpoints to reduce overhead. “Future valuations will hinge on enterprise unit economics, gross margins, and customer retention, forcing severe consolidation among smaller players unable to scale commercial revenue engines or achieve software-like operating leverage,” explains Tripathy. Companies building proprietary tools around smaller language models must prepare for those providers to be absorbed by larger entities or forced out of the market entirely. Designing middleware layers that allow smooth swapping of foundational models is a vital defensive measure against vendor bankruptcy or acquisition. In addition, enterprises should expect more aggressive rate limiting. In a private model, absorbing the compute cost of heavy user requests serves as a loss leader to build market dominance. In a public model, unmetered access destroys gross margins. Businesses will likely see the introduction of complex, tiered pricing structures that penalise erratic workloads and reward predictable, batch-processed data requests. The test for high-capital innovation Anthropic’s journey to the public exchange serves as a barometer for how institutional capital values resource-intensive technology. Samengo-Turner expands on the wider implications for venture-backed companies: “The significance extends well beyond the AI sector. A successful listing could become a reference point for how public markets assess a new generation of technology companies that combine immense capital needs, world-class research talent, and long-term strategic ambitions.” He notes that this event could “encourage more venture-backed technology companies to revisit public markets after a decade in which many of the sector’s biggest growth stories remained private.” If Anthropic successfully sets a public valuation framework, a wave of machine learning companies will likely follow, moving the entire vendor ecosystem toward strict financial compliance and margin protection. “Ultimately, investors will be evaluating more than Anthropic’s prospects,” Samengo-Turner concludes. “They will be testing whether public markets are prepared to support the next generation of technology champions.” See also: Anthropic releases Claude Opus 4.8 Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Anthropic IPO filing marks AI maturing into enterprise utility appeared first on AI News. View the full article

Since its announcement in April this year, the proposed changes to billing methods on GitHub Copilot were a source of much speculation: how much more or less would a pay-a-you-use AI cost an organisation or individual compared to a flat-rate, monthly subscription? Just a day into the changeover to token-based billing for the LLM-based service, software developers and IT departments have been reporting their findings online – and the shortened version is that, as of 1st June 2026, using GitHub Copilot in software development and deployment just got a whole lot more expensive. What are the changes to GitHub Copilot’s charging scheme? Although subscription prices have not changed (Copilot Pro $10 per month, Pro+ $39, Business tier $19 per user, and Enterprise $39 per user per month), the prices now refer to a monthly number of credits that can be spent on the various AI models made available on the GitHub platform. For a typical user, one credit costs a single cent, and depending on the model variant selected at the point of inference, credits are then deducted according to how much silicon effort is expended by the AI. Thus, a Copilot Enterprise user receives 3,900 credits per month ($39), a Copilot Business user receives 1,900 credits ($19). Users will burn up their credits in the form of tokens which are priced differently, according to the power and type of model used. For example, using ChatGPT-5.2, it costs $1.75 per million input tokens (a token can be thought of as nearly-a-word), output tokens cost $14 per million, and cached input (the information held by the LLM to provide ongoing context to a series of queries, for example) are priced at $0.175 per million tokens. When users reach the end of their allotted number of credits, they have the option to buy more. Code completions inside a developer’s IDE (integrated development environment) and ‘next edit’ suggestions will be free, but Code Review processes will cost at the same rates as other GitHub Copilot activities. Will users pay more to use Copilot? Whether or not an average user will end up paying more depends very much on the individual user, and to a certain extent on who you ask. The Comments section of the GitHub Community Discussions page that announced the changes back in April 2026 has many reports of users finding that their credits are being exhausted much more quickly than expected. User ‘rvs99’ said, “My 12% of total AI credits burned like anything for very minor task. I used Claude Sonnet 4.6 as usual and in response it barely updated 2-3 lines in total 6 files which costed like ~$0.35 per line updates.” ‘prhost’ posted a screen-grab of their account dashboard that showed 3,705 credits remaining of an allowance of 7,000 after one day’s use, and stated “It would be easier to shut down the project. [Microsoft] shot themselves in the foot.” User ‘zoomp05’ summarised the tone of most commentators: “The strategy is clear, but it would have been good to say from the beginning, ‘This is a subsidized trial’ or something similar, to promote our tool.” The initial subscription offerings from GitHub, now deprecated, were likely seen by the platform’s owners, Microsoft, as loss leaders. It was immediately apparent that allowing users to burn far more tokens than their subscription value represented was never going to be sustainable. Cursory reading around the internet away from the big model providers’ announcements and posts revealed that, as a business model, subscription-based billing could only be temporary. What is surprising, perhaps, is the surprise of many users that their coding platform is now being billed for at levels in keeping with suppliers’ costs. Running an LLM is not a cheap undertaking, especially considering the additional sums involved in developing new models, post-training, maintenance, data centre construction, future loan repayments, and so on. What businesses might do now Those invested in supporting their development teams with LLM-based coding tools have several options they might consider: Reassess the ROI that AI coding platforms bring, and adjust budget allocations accordingly. Consider which processes of the software development workflow may be cheaper to hand off to AI (junior developer-level code creation, for example), and which are cost sinks (code review, multiple-agent workflows, fast-cadence Actions, etc.). Look for alternative, lower-cost platforms. These fall into three main camps: Open models hosted on-premise. These are not frontier LLMs, and lack many of the features of the coding ‘harnesses’ that professional coding platforms offer. Hosted near-frontier models from LLM providers such as Huawei and Alibaba. ‘Secondary’ coding platforms such as Cursor may offer temporary respite, although be aware that many of the alternatives use frontier, better-known models from OpenAI and Anthropic, and are likely to therefore adopt the same per-use billing as GitHub Copilot. (Image source: Pixabay, under licence.) Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post GitHub Copilot users see token-based price hikes appeared first on AI News. View the full article

Automation is becoming a ******* part of how financial markets are approached, and forex trading is one area where this is becoming easier to notice. As the tech world improves, more traders are looking for ways to stay involved in the market without the need to sit in front of charts for hours at a time. A large part of this thought process comes down to forex robots, which are designed to carry out trades based on a set of pre-made rules. These tools are not new, but they are becoming more refined and easier to use as time goes by. If you are to look at the future of automated trading through the best forex robot reviews, you’ll have a clearer idea of how these systems are being used today and how they may continue to develop over time. How automation use is growing in forex trading Automated trading has been around for a while, but the possibilities available today are more advanced than what traders had access to in the past. Forex robots are able to scan the market, look for specific trade setups, and place trades without a trader ever needing to lift a finger. These systems follow a set of rules that are usually based on technical indicators or past price behaviour. Basically, they are designed to look for patterns and react when certain conditions are met. Some systems are quite basic, and some are built to handle ******* amounts of data and more detailed strategies, so that you don’t have to constantly monitor the market. The growing role of data and AI Artificial intelligence is growing at a fast pace, and playing a more noticeable role in trading these days. Some systems are now able to identify patterns that might not be easy to spot when looking at charts manually. This doesn’t mean that every forex robot is fully driven by AI, but many are now starting to use data in smarter ways. In some cases, systems can adjust how they react based on current market conditions, not following the exact same response every time. This is where FXSentry has become especially useful, as they help break down how different systems work, making it easier for traders to understand what is happening behind the scenes before deciding which one to use. Efficiency and ease of use for traders One of the main reasons that automated trading continues to grow is how it makes trading easier to manage. Certainly not everyone has the time to sit and monitor charts throughout the day, especially in a market that operates almost around the clock. Forex robots run in the background and only jump in when certain conditions are met. This means trades can still be placed even when you’re not actively watching the market, making trading feel more manageable and less overwhelming. Why reviews are important Since there are so many forex robots available, choosing the right one can feel a bit stressful. Each system works differently depending on how it is built and what type of strategy it follows. Reviews can give a clearer picture of how a system operates, how it manages trades, and what kind of results it has produced over time, which makes it easier to compare options and avoid choosing a system without fully understanding how it works first. Important things to keep in mind Even though automated trading can be helpful, it is not without its limits. Markets can change quickly, and a system that performs well under certain conditions may not perform the same way when those conditions change. Forex robots may struggle when something unexpected happens that deviates from their rules. There are also practical factors to think about, like internet connection, platform reliability, and how quickly trades are executed. Because of this, automated systems should be used as support tools not something that replaces decision-making completely. Keeping an eye on performance and making adjustments when needed is still an important part of trading. What the future holds Looking ahead, automated trading could become more advanced as technology continues to improve. Systems may become better at reacting to market conditions and handling more complex data in a way that feels more natural. The future of automated trading will depend not only on how technology improves, but also on how traders choose to apply these tools in a practical and informed way. Wrapping it up Automation is becoming a more common part of forex trading in daily life, helping traders manage their time, follow structured strategies, and stay active in the market without a need for constant monitoring. Exploring the future of automated trading with FXSentry shows how these tools can support a steadier and more manageable approach. When used with a clear understanding of how they work and where their limits are, they can form a useful part of a modern trading setup. The post The future of automated trading with the best forex robot reviews appeared first on AI News. View the full article

A Google Cloud survey found that 90% of developers are already integrating AI into their daily work, and on Steam, 7,818 titles disclosed AI use in 2025 alone, a 681% increase over the previous year. AI in video game development is not a side experiment. It is restructuring the pipeline from concept through launch, and the areas where it is having the most concrete impact are worth examining individually. Smarter NPCs and adaptive gameplay Non-player character behaviour has moved well past scripted decision trees. Ubisoft’s La Forge division developed Ghostwriter, a generative AI tool that produces first-draft NPC dialogue so writers can concentrate on narrative not volume. Large language models now give NPCs genuine memory in sessions and responses that hold up under improvised player input. Alongside this, AI systems monitor player performance in real time to adjust difficulty dynamically, while story engines weave branching subplots on the fly, making each playthrough genuinely distinct. Generative AI and asset creation Andreessen Horowitz has documented cases where concept art generation dropped from three weeks to a single hour once AI tools entered the workflow. Tencent’s Hunyuan3D-PolyGen produces art-grade 3D assets with artists reporting efficiency gains of over 70%, while Meta’s WorldGen can generate a traversable 3D environment from a text prompt in around five minutes, game-engine-ready for Unity and Unreal. Audio is following the same trajectory, with tools like ElevenLabs enabling voice generation and localisation at a speed that traditional recording pipelines cannot match. Quality assurance and playtesting QA is where AI is having a substantial operational impact. EA has deployed reinforcement learning agents to autonomously play and stress-test games, catching edge-case bugs in a wider range of gameplay styles than human testers could cover. Square Enix has announced plans to automate 70% of its QA and debugging using generative AI by 2027, in partnership with the University of Tokyo. The emerging model in the industry is hybrid: AI handles the mechanical volume while human testers focus on judgement-driven issues that automation cannot resolve. Procedural generation and living worlds Modern AI-assisted procedural systems go beyond earlier rule-based approaches by conditioning generation on context. Narrative engines now weave branching subplots that respond to player actions and inferred emotional cues, so each session reflects the shape of an individual playthrough not random variation. Research frameworks like PANGeA are demonstrating that large language models can maintain narrative coherence in dynamically generated content, removing the need for the exhaustive hand-authoring that has traditionally limited branching game stories. AI for browser and web game development Web games are structurally simpler than console or PC titles, HTML5, fast load times, pick-up-and-play mechanics, and that simplicity makes AI tools unusually effective at covering the gap for developers without deep technical or artistic backgrounds. Generative AI can handle concept art and basic asset creation in a fraction of the usual time, while AI-assisted code generation helps less experienced developers get a functional prototype into a browser environment. Tools like FRVR AI let any user generate a playable browser game from a text description alone. Platforms like Poki give those games a natural home: free to play for users, with revenue earned through advertising, making the path from prototype to published title more accessible than it has ever been. The limits and labour questions The expansion has not been frictionless. The flood of low-quality AI-generated titles that hit Steam in 2025 raised real questions about quality floors in an environment where content is cheap to produce. Voice actor unions and writers’ guilds are still negotiating the terms under which AI can generate dialogue or clone voices, and the outcome will shape how studios deploy these tools in character-driven productions. What the evidence so far suggests is that AI in video game development pays for itself when it shortens the distance between a creative intent and a usable output, and studios finding genuine value are putting it precisely where the production bottleneck sits. The post AI in video game development: How artificial intelligence is reshaping the industry appeared first on AI News. View the full article

OpenAI’s latest governance frameworks offer enterprise leaders a structured blueprint for scaling safe and compliant AI deployments globally. The adoption of large language models has steadily progressed towards requiring sustainable, commercial-grade architecture. OpenAI has released its Frontier Governance Framework (FGF), documenting how the organisation addresses systemic risk assessment and mitigation. The framework maps directly to the EU’s General-Purpose AI Code of Practice and California’s Transparency in Frontier AI Act, known as the TFAIA. This publication provides a highly practical template, detailing how internal systems and deployment pipelines can be structured to support high-capability machine learning models securely. Translating these regulatory structures into business strategy begins with understanding defined threat categories. The framework defines systemic risk as foreseeable material risks of severe harm. Specifically, this includes scenarios where a model contributes to greater than 50 fatalities or causes $1 billion in property damages from a single incident. While these scenarios sit at the extreme edge of probability, codifying them allows deployment teams to build appropriate safeguards. By defining boundaries early, enterprises can allocate precise compute resources and engineering hours towards continuous post-deployment monitoring and third-party auditing; ensuring applications remain compliant over their lifecycle. Applying tiered risk evaluations to internal systems OpenAI categorises threats across specific domains: cyber offense, chemical, biological, radiological, and nuclear (CBRN) risks, harmful manipulation, and loss of control. The categorisation system utilises distinct risk tiers to evaluate model capabilities. For example, a Tier 3 cyber offense rating applies to a tool-augmented model capable of identifying and developing functional zero-day exploits of all severity levels in many hardened real-world systems without human intervention. In the CBRN category, a Tier 3 model could enable an expert to develop a highly dangerous novel threat vector, comparable to a CDC Class A biological agent, or autonomously complete the synthesis cycle of a regulated biological threat. Rather than viewing these capabilities purely as hazards, internal security teams can use these tiers to establish defined limits for their proprietary model instances, knowing exactly when a coding assistant or research tool requires heavier oversight. The framework also outlines risks tied to harmful manipulation, described as the purposeful distortion of human behaviour, such as using model capabilities for influence operations or election interference. OpenAI notes that this area remains exploratory and is best addressed through system-level mitigations, like post-deployment monitoring, rather than pre-deployment evaluations. For consumer-facing businesses, this suggests that marketing automation systems using language models simply require real-time content classifiers to ensure they generate objective public messaging. Addressing the risk of humans losing the ability to reliably direct or shut down a system, the framework labels this vector as loss of control. A Tier 2 model in this category demonstrates the capability to reliably evade detection across various evaluation methods, including evading chain of thought monitoring. A Tier 3 model is described as being superior to the most expert humans in executing most complex projects and can operate autonomously for extended, sustained periods of time. It demonstrates highly detailed situational awareness and stealth such that monitoring the model and its chain of thought cannot reliably detect or rule out evasion of human control. By setting these parameters, businesses relying on autonomous agents for supply chain logistics or financial trading have a defined mandate to build deterministic fail-safes and maintain consistent human oversight in automated workflows. Addressing integration challenges and information security OpenAI aligns its internal security with ISO 27001, 27017, 27018, and 27701 standards, alongside SOC 2 Type II evaluations. To protect unreleased model weights, the company employs encryption for data at rest and in transit, multi-factor authentication, and strict multi-party approval protocols. Internal personnel undergo regular training, and model execution occurs in a sandboxed environment with restricted egress by default. When enterprises mirror this setup, they establish a secure baseline for internal operations. Integrating models into proprietary corporate data environments often leads engineering teams to rely on Retrieval-Augmented Generation and dense vector databases. Securing these databases against adversarial prompting or data extraction attempts requires dedicated computational overhead. Every API request passes through security classifiers before hitting the vector database, and the retrieved context is screened before generating a final response. While bridging modern cloud-hosted AI governance structures with older mainframe data silos forces teams to build bespoke, heavily-encrypted middleware, this engineering work results in stable enterprise-ready infrastructure. Maintaining ecosystem compliance and incident response To maintain accurate risk baselines, OpenAI solicits input from external domain experts and independent third-party evaluators. These external experts help stress-test safeguards for models approaching a new risk tier and provide independent opinions to the internal Safety Advisory Group. CDOs within enterprises can similarly benefit from external auditing retainers to independently verify that their localised model deployments remain within acceptable risk thresholds. Connecting to the broader regulatory ecosystem, external reporting dictates the ongoing operational cadence. OpenAI documents its mitigation results in a Safety and Security Model Report. Under the EU AI Act provisions, the company commits to evaluating whether to update these reports for its most capable models every six months. Updates to the reports are considered required if a model’s capabilities materially change through post-training or if integrations into internal systems increase risk. The responsibility for EU compliance rests with OpenAI Ireland Limited, while OpenAI OpCo LLC manages obligations under the TFAIA in the US. To manage sudden software anomalies, OpenAI utilises an AI Safety Incident Response Plan, abbreviated as the AIRP. This plan dictates procedures for triage, investigation, and external reporting of severe safety incidents. Potential incidents are flagged through automated monitoring, employee escalation, or end-user feedback. Once flagged, response teams investigate the root cause, scope, and impact, taking action to mitigate and contain the event. Enterprise leaders can easily mirror these response mechanisms; establishing parallel internal response units capable of adjusting anomalous API behaviour proactively. Within OpenAI, updates to the framework can be proposed by various leaders, including the Head of Safety Systems, CISO, and General Counsel. The company conducts a formal Framework Assessment at least once every 12 months; evaluating changes in law, new model capabilities, and industry standards. The integration of advanced computational models remains a viable path to corporate efficiency, and adopting these frameworks ensures the internal architecture is well-prepared to handle modern compliance demands securely. See also: Anthropic releases Claude Opus 4.8 Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Scaling safe enterprise AI with OpenAI governance frameworks appeared first on AI News. View the full article

Anthropic has released Claude Opus 4.8, an upgrade to Claude Opus 4.7 that the company says brings improved results for coding, agent work, reasoning, and knowledge work. The platform can be used through claude.ai, Claude Code and the Claude API, with the API name claude-opus-4-8. The company has also altered some of the details of its product line-up. Users of claude.ai and Cowork can set the amount of effort Claude applies to a response – essentially, affecting the number of tokens the model will burn. Claude Code also has dynamic workflows, a feature that plans work, runs parallel sub-agents, verifies outputs and reports back to the user. Finally, the Messages API accepts live changes to the messages array, which Anthropic says lets developers update instructions during a task without breaking prompt cache use or needing a separate user turn. Anthropic said the price for use of Claude Opus 4.8 when not in ‘fast’ mode will remain at $5 per million input and $25 per million output tokens, while fast mode costs $10 per million input tokens and $50 per million output. Fast mode for Opus 4.8 works at 2.5x, the company’s announcement post states. The company has positioned Opus 4.8 as designed for coding and agentic workflows in coding, where the model can use tools inside a context and check its own work. It says Opus 4.8 improves on Opus 4.7 on benchmarks for coding, agent skills, reasoning, and office work. There is a System Card that can be examined for further subjective detail. Anthropic’s announcement cites several companies that have tested the platform before its wider release, including those operating in software development, law, finance, and research. Several testers commented on the platform’s agentic workflows, with one noting a cost parity with GPT-5.5 when running its internal benchmark tests. A comment from CursorBench said Opus 4.8 used fewer tool steps to achieve the same level of output. Anthropic says Opus 4.8 is less likely than its 4.7 predecessor to pass flawed code without comment, which it describes four times less likely. It says the platform showed lower rates of deception or the tendency to go along with misuse than Opus 4.7 and is comparable in this regard as those exhibited by Claude Mythos Preview. Effort control helps users to manage any trade-off between quality, speed, and token burn rates. Opus 4.8 defaults to high effort but on coding tasks, the company said the higher default only uses the type of token numbers of Opus 4.7, but performs better. Users can opt for ‘xhigh’ for work that needs more computation. Anthropic said it has increased Claude Code rate limits to support the resulting higher token use. Dynamic workflows in Claude Code are designed for large codebases, and can migrate codebases of hundreds of thousands of lines. These features are currently in research preview and are available on the Enterprise, Team, and Max plans. The Messages API updates instructions during an agent’s run, with edits inside the messages array being used, for example, to update permissions, change token budgets or context while agents continue their work. Anthropic also used the release to suggest it’s developing models that provide current levels of ability at less cost to the user, and will release a class of model that’s better than the current Opus platform. Its roadmap includes Project Glasswing, under which a group of organisations is using Claude Mythos Preview for cybersecurity scanning. Anthropic said models at that capability level require stronger safeguards before release to all customers. It expects to bring ‘Mythos-class’ models to customers in the coming weeks. The additional controls in 4.8 will expose the cost and effort trade-offs to users as the company transitions to token-based billing from subscription tiers. (Image source: Pixabay, under licence.) Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Anthropic releases Claude Opus 4.8 appeared first on AI News. View the full article

Google Pay is overhauling its payment infrastructure for an impending wave of transactions from AI agents. The latest updates introduce the Universal Commerce Protocol and a new server architecture, positioning Google Pay as a central clearinghouse for purchases executed by autonomous agents rather than human users. AI agents – designed to perform tasks like booking flights or ordering supplies – cannot effectively navigate the multi-step, visually-oriented checkout pages built for human interaction. Google is attempting to replace this UI-dependent model with a stable, API-driven backend for machines. This restructuring of Google Pay introduces several components: Universal Commerce Protocol (UCP): This is a new specification intended to standardise how AI agents communicate with payment and merchant systems. It creates a common language for initiating transactions, confirming inventory, and handling fulfillment details. The objective is to eliminate the need for developers to build bespoke integrations for every merchant or payment provider an agent might interact with. New Merchant Commerce Platform (MCP) server: Google is deploying a new server-side system to act as an intermediary. This MCP server manages merchant integrations and analyses transaction trends. For developers building agents, it abstracts away the complexity of the commerce backend. For Google, it centralises a vast amount of transactional data from agent-driven activities. Dynamic callbacks for Android native: To facilitate more complex checkouts, Google is enabling dynamic callbacks within its Android Pay API. This allows for real-time adjustments to an order (e.g. updating shipping costs based on a new address or recalculating tax) without forcing the user or agent to restart the entire process. It makes the transaction flow more resilient to mid-process changes. Expanded WebView support: The company is extending payment support within WebViews. This is a critical detail, as it allows transactions to be completed inside third-party applications, particularly social media platforms where conversational commerce is expected to increase. Agents operating within these environments can now execute payments natively. Realities of machine-to-machine commerce The concept of a customer journey, once defined by clicks and page views, now extends to an agent’s ability to parse product data and execute a transaction via an API. Marketing leaders now have to consider “search engine optimisation” for machines. Product information, pricing, and availability will need to be presented as machine-readable data, not just persuasive copy for a human audience. If an AI agent cannot parse your inventory data to make a purchasing decision, your business becomes invisible in this new commercial channel. The introduction of the MCP server also raises questions about data governance and vendor dependency. By routing transactions through its platform, Google gains a privileged view of commerce trends driven by AI agents. CIOs must assess the long-term implications of building reliance on a proprietary protocol and a centralised data aggregation point. The convenience of a universal standard comes with the strategic cost of platform lock-in. New architectures for security and trust Authorising transactions initiated by an autonomous agent presents a new set of security challenges. A faulty or malicious agent could execute unauthorised purchases at scale. Google’s answer is the introduction of cross-device biometric authentication. This mechanism allows an AI agent to programmatically request human verification for a transaction. A user could receive a prompt on their phone to approve a purchase an agent has arranged on their laptop. This approach establishes a “human-in-the-loop” security model for high-value or sensitive transactions. It provides a necessary kill-switch and audit trail for agent activities. Defining the policies for when an agent can act autonomously versus when it must seek human approval becomes a new area of corporate governance. These rules will need to be encoded into the agent’s operational logic, creating a direct link between business policy and software behaviour. These latest updates to Google Pay are an early but concrete signal of the architectural changes required to support a machine-driven economy. Enterprises that continue to view their digital presence as a collection of websites for human consumption will be unprepared for this next phase of commerce. See also: Google folds Display Ads into AI-first Demand Gen platform Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Google Pay preps for AI agents with Universal Commerce Protocol appeared first on AI News. View the full article

NBA Commissioner Adam Silver said the league plans to introduce an automated system for certain officiating decisions, including out-of-bounds calls. The system would use AI and cameras placed around the court to determine possession. Silver compared the approach to Hawk-Eye, the tracking technology used for line calls in tennis. Disputed call preceded Silver’s comments Silver’s appearance came after a disputed call in Game 5 of the Western Conference finals between the Oklahoma City Thunder and San Antonio Spurs. Late in the third quarter, Spurs centre Victor Wembanyama was ruled to have touched the ball last on an out-of-bounds play. The replay showed the ball had bounced off the foot of Thunder forward Chet Holmgren. The call stood after the officials conferred. The call drew attention after Oklahoma City took a 3-2 lead in the series. Silver said the NBA eventually intends to remove that category of objective calls from on-court officials. NBA partnership with Hawk-Eye started in 2023 The NBA previously announced work with Sony’s Hawk-Eye Innovations. In 2023, the league said it had entered a multi-year partnership to deploy 3D optical tracking technology. The partnership followed several years of testing at Summer League and NBA arenas. The NBA said the system was designed to track the ball and player movement in three dimensions at sub-second latency. The league also named out-of-bounds and goaltending as possible future use cases for automated officiating. Silver referred to out-of-bounds calls during the ESPN appearance. Automated officiating systems are used in defined call categories in other sports. Tennis uses electronic line calling, while FIFA has used semi-automated offside technology. MLB is introducing an automated ******-and-strikes challenge system in 2026. “We’re going to move to a system like that where that whole category of calls will be automatic,” Silver said. He said the system would determine possession immediately, whether the ball belongs to the Lakers, Knicks, Thunder, Spurs, or another team. Silver said the system would reduce the need for challenges on those calls. Coach’s Challenge covers out-of-bounds reviews Under current NBA rules, a Coach’s Challenge is the only way to trigger replay review of an out-of-bounds violation at any point during a game. Each team starts with one challenge and receives a second only if the first challenge is successful. The NBA also expanded the Coach’s Challenge rule for the 2024–25 season. The change allows officials to review whether a foul should have been called during certain out-of-bounds reviews. Silver said the technology would allow games to continue without stoppages for that category of decision. “It’ll be instantaneous, it’ll be automatic. Just play on,” he said. The NBA has already expanded its use of replay review and centralised officiating support. The league operates a Replay Center in Secaucus, New Jersey. According to the NBA, all 30 arenas are connected to the facility, which has 94 HD monitors, 23 workstations, and supports reviews across 15 instant replay triggers. Referees remain responsible for fouls Silver said referees would remain responsible for calls that require judgment, especially those involving contact and fouls. He said contact occurs on many plays, but officials still need to decide whether the contact affected a player’s movement or ability to continue the play. “There’s often contact on every play,” Silver said. “It doesn’t mean there’s a foul.” Silver did not give a specific timeline for introducing the automated system. He said the league expects to move in that direction “fairly quickly.” (Photo by JC Gellidon) See also: Autonomous AI systems test governance in physical environments Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post NBA plans AI system for automatic out-of-bounds calls appeared first on AI News. View the full article

Google is folding Display Ads into its AI-powered Demand Gen platform, marking the end of a long-standing digital advertising model. The Google Display Network (GDN) has been a staple of the open internet for almost twenty years. Marketers previously relied on its predictable framework to target placements, bid on audiences, and A/B test static creative across news sites and blogs. That familiar setup is changing and requires marketing teams to move away from manual campaign controls and rely on Google’s AI. Google describes this change as a natural progression and presents it as a method for advertisers to reach visual platforms like YouTube, Discover, and Gmail through one consolidated campaign. Traditional banner ads are facing increased competition from the full-screen video formats of platforms like TikTok and Instagram. In response, Google’s Demand Gen uses an automated system to generate and develop customer interest before a search query is ever entered. Demand Gen functions differently from the traditional GDN. Instead of having advertisers select specific websites or adjust audience segments, the platform requires business goals and a collection of creative assets. Marketers upload images, video clips, and headlines, which Google’s AI then tests in various combinations. The system serves these as in-stream video ads, YouTube Shorts, or interactive Discover posts, using predictive models to determine format, placement, and audience. This transition requires changes to creative production. Demand Gen relies on a continuous supply of diverse, format-agnostic content. Creative teams are now tasked with providing the raw assets that Google’s AI assembles dynamically, shifting the traditional agency workflow toward higher-volume content creation. Trading granularity for automation Google is betting that machine learning will beat human intuition at scale, effectively forcing the industry’s hand. Consolidating Display into this AI-centric model removes the temptation for teams to cling to manual methods. Advertisers must adopt the AI-first approach or risk losing visibility on valuable digital real estate. Long-standing metrics like click-through rate (CTR) and cost-per-click (CPC) are now losing much of their meaning. Judging the success of a single creative or placement becomes nearly impossible when an AI optimises for conversions or brand lift simultaneously across multiple formats and platforms. Instead, reporting must elevate to track broader business outcomes: customer acquisition cost, return on ad spend, and influence on the overall purchase journey. This requires tighter integration between advertising platforms and a company’s core business intelligence systems. Without accurate, real-time conversion data, the AI flies blind. For many enterprises, this dependency exposes critical weaknesses in their data infrastructure. A multi-million-pound Demand Gen budget could easily hinge on the quality of a single API connection to a CRM or e-commerce backend that are often built for entirely different purposes. Meta pushes a similar agenda with its Advantage+ campaigns, leveraging AI to automate targeting, creative, and placement across its ecosystem. The industry is clearly shifting from a model of renting ad space to one of commissioning AI agents to hunt down customers. Marketing leaders no longer have a choice about ceding control to AI; the focus is on how they adapt their teams, technology, and strategy. See also: Musk and Zuckerberg convinced Trump to scrap AI executive order Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Google folds Display Ads into AI-first Demand Gen platform appeared first on AI News. View the full article