ChatGPT

Apple is planning to use a custom version of Google’s Gemini model to support a major upgrade to Siri, according to Bloomberg’s Mark Gurman. The company may pay Google about $1 billion each year for access to technology that can create summaries and handle planning tasks. Bloomberg says Apple will run the custom model on its Private Cloud Compute servers, while still relying on its own systems for some parts of Siri. Gurman reports that the Gemini model uses 1.2 trillion parameters, far more than the 150 billion parameters behind the current cloud-based version of Apple Intelligence. Apple is preparing to spend about $1 billion each year on a powerful Google-built artificial intelligence model with 1.2 trillion parameters, according to people familiar with the matter, as reported by Bloomberg. The system is expected to play a central role in a major update to Siri, a project the company has been working toward for years. After months of testing, Apple and Google are close to a deal that would give Apple access to the technology. The people discussing the plans asked not to be named because the talks are private. Apple is turning to Google to help rebuild Siri’s core technology, laying the groundwork for a broad refresh of features planned for next year. The size of Google’s model would far exceed the AI systems Apple uses today. Apple tested other outside options — including Google’s Gemini, OpenAI’s ChatGPT, and Anthropic’s Claude — before deciding to move forward with Google earlier this year. The goal is to rely on Gemini as a temporary solution until Apple’s own work reaches the same level. The updated Siri is planned for release next spring, Bloomberg reported. Because months remain before launch, parts of the plan could still change. Apple and Google declined to comment. Shares of both companies briefly rose after the news surfaced Wednesday. Apple’s stock gained less than 1% to $271.70, while Alphabet climbed as much as 3.2% to $286.42. The custom Gemini model would be a major jump from the 150 billion parameter system Apple currently uses in the cloud for Apple Intelligence. The move is meant to increase Siri’s ability to process complex tasks and understand context at a deeper level. The work is known internally as Glenwood and is led by Vision Pro headset creator Mike Rockwell and software chief Craig Federighi. The refreshed voice assistant, set to appear in iOS 26.4, is code-named Linwood. Under the deal, Google’s model will support Siri’s summarizer and planner features — the parts that help the assistant understand information and decide on action steps. Apple’s own models will still handle some tools and responses. The model will run on Apple’s Private Cloud Compute servers, keeping user data isolated from Google’s systems. Apple already set aside server hardware for the effort to support Siri’s new features. Although the partnership is large, Apple is not expected to promote it to consumers. Google will act as a quiet technology provider, unlike the visible search agreement inside Safari. Siri’s improvements will likely appear without Google branding. This deal is separate from earlier talks about placing Gemini directly inside Siri as a chatbot. Those conversations nearly turned into a product in both 2024 and again earlier this year, but never moved forward. The new agreement also does not place Google AI search features inside Apple’s operating systems, leaving Siri’s search behavior unchanged. During Apple’s most recent earnings call, Chief Executive Officer Tim Cook said Siri may add more chatbot options in the future, beyond the current ChatGPT choice. Apple continues to look for ways to expand Siri without relying on one provider. Other companies are also adopting Gemini. Snap and several major firms are building products using Google’s Vertex AI platform. For Apple, the move reflects how far behind it has fallen in AI — and how willing the company is to use outside tools to improve Siri. Even so, Apple does not plan to use Gemini forever. The company has lost AI engineers in recent years, including the head of its models team, but Apple’s leadership still wants to develop its own technology and eventually replace Google’s system inside Siri. Apple’s internal team is building its own cloud-based model with up to 1 trillion parameters, which could be ready for consumer use as early as next year. That work is expected to support Siri’s growth in the long run. Executives believe they can match Google’s quality over time. But Google continues to improve Gemini, making the gap harder to close. Its 2.5 Pro version ranks near the top of most large language model comparisons, which affects how Apple plans Siri’s updates. Apple also wants to bring Apple Intelligence and the updated Siri to China. Because Google services are banned in the country, the system used there will not rely on Gemini. Instead, Apple plans to use its own models along with a content filter built by Alibaba Group Holding Ltd. That tool would adjust responses to meet government requirements. Apple has also explored a partnership with Baidu Inc. for AI features in the ******** market, Bloomberg reported earlier this year. (Photo by omid armin) See also: Inside Tim Cook’s push to get Apple back in the AI race 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 Apple plans big Siri update with help from Google AI appeared first on AI News. View the full article

Enterprise leaders grappling with the steep costs of deploying AI models could find a reprieve thanks to a new architecture design. While the capabilities of generative AI are attractive, their immense computational demands for both training and inference result in prohibitive expenses and mounting environmental concerns. At the centre of this inefficiency is the models’ “fundamental bottleneck” of an autoregressive process that generates text sequentially, token-by-token. For enterprises processing vast data streams, from IoT networks to financial markets, this limitation makes generating long-form analysis both slow and economically challenging. However, a new research paper from Tencent AI and Tsinghua University proposes an alternative. A new approach to AI efficiency The research introduces Continuous Autoregressive Language Models (CALM). This method re-engineers the generation process to predict a continuous vector rather than a discrete token. A high-fidelity autoencoder “compress[es] a chunk of K tokens into a single continuous vector,” which holds a much higher semantic bandwidth. Instead of processing something like “the”, “cat”, “sat” in three steps, the model compresses them into one. This design directly “reduces the number of generative steps,” attacking the computational load. The experimental results demonstrate a better performance-compute trade-off. A CALM AI model grouping four tokens delivered performance “comparable to strong discrete baselines, but at a significantly lower computational cost” for an enterprise. One CALM model, for instance, required 44 percent fewer training FLOPs and 34 percent fewer inference FLOPs than a baseline Transformer of similar capability. This points to a saving on both the initial capital expense of training and the recurring operational expense of inference. Rebuilding the toolkit for the continuous domain Moving from a finite, discrete vocabulary to an infinite, continuous vector space breaks the standard LLM toolkit. The researchers had to develop a “comprehensive likelihood-free framework” to make the new model viable. For training, the model cannot use a standard softmax layer or maximum likelihood estimation. To solve this, the team used a “likelihood-free” objective with an Energy Transformer, which rewards the model for accurate predictions without computing explicit probabilities. This new training method also required a new evaluation metric. Standard benchmarks like Perplexity are inapplicable as they rely on the same likelihoods the model no longer computes. The team proposed BrierLM, a novel metric based on the Brier score that can be estimated purely from model samples. Validation confirmed BrierLM as a reliable alternative, showing a “Spearman’s rank correlation of -0.991” with traditional loss metrics. Finally, the framework restores controlled generation, a key feature for enterprise use. Standard temperature sampling is impossible without a probability distribution. The paper introduces a new “likelihood-free sampling algorithm,” including a practical batch approximation method, to manage the trade-off between output accuracy and diversity. Reducing enterprise AI costs This research offers a glimpse into a future where generative AI is not defined purely by ever-larger parameter counts, but by architectural efficiency. The current path of scaling models is hitting a wall of diminishing returns and escalating costs. The CALM framework establishes a “new design axis for LLM scaling: increasing the semantic bandwidth of each generative step”. While this is a research framework and not an off-the-shelf product, it points to a powerful and scalable pathway towards ultra-efficient language models. When evaluating vendor roadmaps, tech leaders should look beyond model size and begin asking about architectural efficiency. The ability to reduce FLOPs per generated token will become a defining competitive advantage, enabling AI to be deployed more economically and sustainably across the enterprise to reduce costs—from the data centre to data-heavy edge applications. See also: Flawed AI benchmarks put enterprise budgets at risk 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 Expo, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Keep CALM: New model design could fix high enterprise AI costs appeared first on AI News. View the full article

Boards of directors are pressing for productivity gains from large-language models and AI assistants. Yet the same features that makes AI useful – browsing live websites, remembering user context, and connecting to business apps – also expand the cyber attack surface. Tenable researchers have published a set of vulnerabilities and attacks under the title “HackedGPT”, showing how indirect prompt injection and related techniques could enable data exfiltration and malware persistence. Some issues have been remediated, while others reportedly remain exploitable at the time of the Tenable disclosure, according to an advisory issued by the company. Removing the inherent risks from AI assistants’ operations requires governance, controls, and operating methods that treat AI as a user or device, to the extent that the technology should be subject to strict audit and monitoring The Tenable research shows the failures that can turn AI assistants into security issues. Indirect prompt injection hides instructions in web content that the assistant reads while browsing, instructions that trigger data access the user never intended. Another vector involves the use of a front-end query that seeds malicious instructions. The business impact is clear, including the need for incident response, legal and regulatory review, and steps taken to reduce reputational harm. Research already exists that shows assistants can leak personal or sensitive information through injection techniques, and AI vendors and cybersecurity experts have to patch issues as they emerge. The pattern is familiar to anyone in the technology industry: as features expand, so do failure modes. Treating AI assistants as live, internet-facing applications – not productivity drivers – can improve resilience. How to govern AI assistants, in practice 1) Establish an AI system registry Inventory every model, assistant, or agent in use – in public cloud, on-premises, and software-as-a-service, in line with the NIST AI RMF Playbook. Record owner, purpose, capabilities (browsing, API connectors) and data domains accessed. Even without this AI asset list, “shadow agents” can persist with privileges no one tracks. Shadow AI – at one stage encouraged by the likes of Microsoft, who encouraged users to deploy home Copilot licences at work – is a significant threat. 2) Separate identities for humans, services, and agents Identity and access management conflate user accounts, service accounts, and automation devices. Assistants that access websites, call tools, and write data need distinct identities and be subject to zero-trust policies of least-privilege. Mapping agent-to-agent chains (who asked whom to do what, over which data, and when) is a bare minimum crumb trail that may ensure some degree of accountability. It’s worth noting that agentic AI is susceptible to ‘creative’ output and actions, yet unlike human staff, are not constrained by disciplinary policies. 3) Constrain risky features by context Make browsing and independent actions taken by AI assistants opt-in per use case. For customer-facing assistants, set short retention times unless there’s a strong reason and a lawful basis otherwise. For internal engineering, use AI assistants but only in segregated projects with strict logging. Apply data-loss-prevention to connector traffic if assistants can reach file stores, messaging, or e-mail. Previous plugin and connector issues demonstrate how integrations increase exposure. 4) Monitor like any internet-facing app Capture assistant actions and tool calls as structured logs. Alert on anomalies: sudden spikes in browsing to unfamiliar domains; attempts to summarise opaque code blocks; unusual memory-write bursts; or connector access outside policy boundaries. Incorporate injection tests into pre-production checks. 5) Build the human muscle Train developers, cloud engineers, and analysts to recognise injection symptoms. Encourage users to report odd behaviour (e.g., an assistant unexpectedly summarising content from a site they didn’t open). Make it normal to quarantine an assistant, clear memory, and rotate its credentials after suspicious events. The skills gap is real; without upskilling, governance will lag adoption. Decision points for IT and cloud leaders Question Why it matters Which assistants can browse the web or write data? Browsing and memory are common injection and persistence paths; constrain per use case. Do agents have distinct identities and auditable delegation? Prevents “who did what?” gaps when instructions are seeded indirectly. Is there a registry of AI systems with owners, scopes, and retention? Supports governance, right-sizing of controls, and budget visibility. How are connectors and plugins governed? Third-party integrations have a history of security issues; apply least privilege and DLP. Do we test for 0-click and 1-click vectors before go-live? Public research shows both are feasible via crafted links or content. Are vendors patching promptly and publishing fixes? Feature velocity means new issues will appear; verify responsiveness. Risks, cost visibility, and the human factor Hidden cost: assistants that browse or retain memory consume compute, storage, and egress in ways finance teams and those monitoring per-cycle Xaas use may not have modelled. A registry and metering reduce surprises. Governance gaps: audit and compliance frameworks built for human users won’t automatically capture agent-to-agent delegation. Align controls according to OWASP LLM risks and NIST AI RMF categories. Security risk: indirect prompt injection can be invisible to users, passed from media, text or code formatting, as shown by research. Skills gap: many teams haven’t yet merged AI/ML and cybersecurity practices. Invest in training that covers assistant threat-modelling and injection testing. Evolving posture: expect a cadence of new flaws and fixes. OpenAI’s remediation of a zero-click path in late 2025 is a reminder that vendor posture changes quickly and needs verification. Bottom line The lesson for executives is simple: treat AI assistants as powerful, networked applications with their own lifecycle and a propensity for both being the subject of attack and for taking unpredictable action. Put a registry in place, separate identities, constrain risky features by default, log everything meaningful, and rehearse containment. With these guardrails in place, agentic AI is more likely to deliver measurable efficiency and resilience – without quietly becoming your newest breach vector. (Image source: “The Enemy Within Unleashed” by aha42 | tehaha is licensed under CC BY-NC 2.0.) 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 The enemy within: AI as the attack surface appeared first on AI News. View the full article

A new academic review suggests AI benchmarks are flawed, potentially leading an enterprise to make high-stakes decisions on “misleading” data. Enterprise leaders are committing budgets of eight or nine figures to generative AI programmes. These procurement and development decisions often rely on public leaderboards and benchmarks to compare model capabilities. A large-scale study, ‘Measuring what Matters: Construct Validity in Large Language Model Benchmarks,’ analysed 445 separate LLM benchmarks from leading AI conferences. A team of 29 expert reviewers found that “almost all articles have weaknesses in at least one area,” undermining the claims they make about model performance. For CTOs and Chief Data Officers, it strikes at the heart of AI governance and investment strategy. If a benchmark claiming to measure ‘safety’ or ‘robustness’ doesn’t actually capture those qualities, an organisation could deploy a model that exposes it to serious financial and reputational risk. The ‘construct validity’ problem The researchers focused on a core scientific principle known as construct validity. In simple terms, this is the degree to which a test measures the abstract concept it claims to be measuring. For example, while ‘intelligence’ cannot be measured directly, tests are created to serve as measurable proxies. The paper notes that if a benchmark has low construct validity, “then a high score may be irrelevant or even misleading”. This problem is widespread in AI evaluation. The study found that key concepts are often “poorly defined or operationalised”. This can lead to “poorly supported scientific claims, misdirected research, and policy implications that are not grounded in robust evidence”. When vendors compete for enterprise contracts by highlighting their top scores on benchmarks, leaders are effectively trusting that these scores are a reliable proxy for real-world business performance. This new research suggests that trust may be misplaced. Where the enterprise AI benchmarks are failing The review identified systemic failings across the board, from how benchmarks are designed to how their results are reported. Vague or contested definitions: You cannot measure what you cannot define. The study found that even when definitions for a phenomenon were provided, 47.8 percent were “contested,” addressing concepts with “many possible definitions or no clear definition at all”. The paper uses ‘harmlessness’ – a key goal in enterprise safety alignment – as an example of a phenomenon that often lacks a clear, agreed-upon definition. If two vendors score differently on a ‘harmlessness’ benchmark, it may only reflect two different, arbitrary definitions of the term, not a genuine difference in model safety. Lack of statistical rigour: Perhaps most alarming for data-driven organisations, the review found that only 16 percent of the 445 benchmarks used uncertainty estimates or statistical tests to compare model results. Without statistical analysis, it’s impossible to know if a 2 percent lead for Model A over Model B is a genuine capability difference or simple random chance. Enterprise decisions are being guided by numbers that would not pass a basic scientific or business intelligence review. Data contamination and memorisation: Many benchmarks, especially those for reasoning (like the widely used GSM8K), are undermined when their questions and answers appear in the model’s pre-training data. When this happens, the model isn’t reasoning to find the answer; it’s simply memorising it. A high score may indicate a good memory, not the advanced reasoning capability an enterprise actually needs for a complex task. The paper warns this “undermine the validity of the results” and recommends building contamination checks directly into the benchmark. Unrepresentative datasets: The study found that 27 percent of benchmarks used “convenience sampling,” such as reusing data from existing benchmarks or human exams. This data is often not representative of the real-world phenomenon. For example, the authors note that reusing questions from a “calculator-free exam” means the problems use numbers chosen to be easy for basic arithmetic. A model might score well on this test, but this score “would not predict performance on larger numbers, where LLMs struggle”. This creates a critical blind spot, hiding a known model weakness. From public metrics to internal validation For enterprise leaders, the study serves as a strong warning: public AI benchmarks are not a substitute for internal and domain-specific evaluation. A high score on a public leaderboard is not a guarantee of fitness for a specific business purpose. Isabella Grandi, Director for Data Strategy & Governance, at NTT DATA ***&I, commented: “A single benchmark might not be the right way to capture the complexity of AI systems, and expecting it to do so risks reducing progress to a numbers game rather than a measure of real-world responsibility. What matters most is consistent evaluation against clear principles that ensure technology serves people as well as progress. “Good methodology – as laid out by ISO/IEC 42001:2023 – reflects this balance through five core principles: accountability, fairness, transparency, security and redress. Accountability establishes ownership and responsibility for any AI system that is deployed. Transparency and fairness guide decisions toward outcomes that are ethical and explainable. Security and privacy are non-negotiable, preventing misuse and reinforcing public trust. Redress and contestability provide a vital mechanism for oversight, ensuring people can challenge and correct outcomes when necessary. “Real progress in AI depends on collaboration that brings together the vision of government, the curiosity of academia and the practical drive of industry. When partnerships are underpinned by open dialogue and shared standards take hold, it builds the transparency needed for people to instil trust in AI systems. Responsible innovation will always rely on cooperation that strengthens oversight while keeping ambition alive.” The paper’s eight recommendations provide a practical checklist for any enterprise looking to build its own internal AI benchmarks and evaluations, aligning with the principles-based approach. Define your phenomenon: Before testing models, organisations must first create a “precise and operational definition for the phenomenon being measured”. What does a ‘helpful’ response mean in the context of your customer service? What does ‘accurate’ mean for your financial reports? Build a representative dataset: The most valuable benchmark is one built from your own data. The paper urges developers to “construct a representative dataset for the task”. This means using task items that reflect the real-world scenarios, formats, and challenges your employees and customers face. Conduct error analysis: Go beyond the final score. The report recommends teams “conduct a qualitative and quantitative analysis of common failure modes”. Analysing why a model fails is more instructive than just knowing its score. If its failures are all on low-priority, obscure topics, it may be acceptable; if it fails on your most common and high-value use cases, that single score becomes irrelevant. Justify validity: Finally, teams must “justify the relevance of the benchmark for the phenomenon with real-world applications”. Every evaluation should come with a clear rationale explaining why this specific test is a valid proxy for business value. The race to deploy generative AI is pushing organisations to move faster than their governance frameworks can keep up. This report shows that the very tools used to measure progress are often flawed. The only reliable path forward is to stop trusting generic AI benchmarks and start “measuring what matters” for your own enterprise. See also: OpenAI spreads $600B cloud AI bet across AWS, Oracle, Microsoft 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 Expo, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Flawed AI benchmarks put enterprise budgets at risk appeared first on AI News. View the full article

Align Technology, a medical device company that designs, manufactures, and sells the Invisalign system of clear aligners, exocad CAD/CAM software, and iTero intra-oral scanners, has unveiled ClinCheck Live Plan, a new feature in its Invisalign digital dental treatment planning. ClinCheck Live Plan is designed to automate the creation of an initial Invisalign treatment plan that’s ready for a practitioner to review and approve, cutting treatment planning cycles from days down to just 15 minutes. The goal is to help patients get the treatment they need faster. The latest plan follows Align’s range of new treatment planning tools and automation features launched in recent years, like cloud-based ClinCheck Pro 6.0 software, the automated Invisalign Personalised Plan templates, and the one-page Flex Rx prescription form for simplified workflows. Each new feature has been designed to improve consistency, dentist control, and speed. Built on Align’s data and algorithms, ClinCheck Live Plan has been in development for decades, with insights from dentists and orthodontists who have treated over 21 million Invisalign patients globally. Dentists will be able to create and adjust treatment plans and, once an eligible case has been submitted using the Flex Rx system, receive a personalised ClinCheck treatment plan in approximately 15 minutes. Invisalign specialists can review their patients’ teeth and how they plan to adjust them, helping improve service while the patient is present. Once an Invisalign clinician submits a new case with an iTero intra-oral scan and a completed Flex Rx prescription, the ClinCheck Live Plan system makes a 3D plan. Ultimately, a faster process should help clinics operate more efficiently and enhance their patients’ experiences. Invisalign-trained specialists that currently use the ClinCheck preferences template and Flex Rx form will gain access to ClinCheck Live Plan when it becomes available in their region. A worldwide rollout of the plan is set to start in the first quarter of 2026. (Image source: “Visiting the dentist in SL” by Daniel Voyager is licensed under CC BY 2.0.) 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 ClinCheck Live brings AI planning to Invisalign dental treatments appeared first on AI News. View the full article

OpenAI is on a spending spree to secure its AI compute supply chain, signing a new deal with AWS as part of its multi-cloud strategy. The company recently ended its exclusive cloud-computing partnership with Microsoft. It has since allocated a reported $250 billion back to Microsoft, $300 billion to Oracle, and now, $38 billion to Amazon Web Services (AWS) in a new multi-year pact. This $38 billion AWS deal, while the smallest of the three, is part of OpenAI’s diversification plan. For industry leaders, OpenAI’s actions show that access to high-performance GPUs is no longer an on-demand commodity. It is now a scarce resource requiring massive long-term capital commitment. The AWS agreement provides OpenAI with access to hundreds of thousands of NVIDIA GPUs, including the new GB200s and GB300s, and the ability to tap tens of millions of CPUs. This mighty infrastructure is not just for training tomorrow’s models; it’s needed to run the massive inference workloads of today’s ChatGPT. As OpenAI co-founder and CEO Sam Altman stated, “scaling frontier AI requires massive, reliable compute”. This spending spree is forcing a competitive response from the hyperscalers. While AWS remains the industry’s largest cloud provider, Microsoft and Google have recently posted faster cloud-revenue growth, often by capturing new AI customers. This AWS deal is a plain attempt to secure a cornerstone AI workload and prove its large-scale AI capabilities, which it claims include running clusters of over 500,000 chips. AWS is not just providing standard servers. It is building a sophisticated, purpose-built architecture for OpenAI, using EC2 UltraServers to link the GPUs for the low-latency networking that large-scale training demands. “The breadth and immediate availability of optimised compute demonstrates why AWS is uniquely positioned to support OpenAI’s vast AI workloads,” said Matt Garman, CEO of AWS. But “immediate” is relative. The full capacity from OpenAI’s latest cloud AI deal will not be fully deployed until the end of 2026, with options to expand further into 2027. This timeline offers a dose of realism for any executive planning an AI rollout: the hardware supply chain is complex and operates on multi-year schedules. What, then, should enterprise leaders take from this? First, the “build vs. buy” debate for AI infrastructure is all but over. OpenAI is spending hundreds of billions to build on top of rented hardware. Few, if any, other companies can or should follow suit. This pushes the rest of the market firmly toward managed platforms like Amazon Bedrock, Google Vertex AI, or IBM watsonx, where the hyperscalers absorb this infrastructure risk. Second, the days of single-cloud sourcing for AI workloads may be numbered. OpenAI’s pivot to a multi-provider model is a textbook case of mitigating concentration risk. For a CIO, relying on one vendor for the compute that runs a core business process is becoming a gamble. Finally, AI budgeting has left the realm of departmental IT and entered the world of corporate capital planning. These are no longer variable operational expenses. Securing AI compute is now a long-term financial commitment, much like building a new factory or data centre. See also: Qualcomm unveils AI data centre chips to crack the Inference market 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 Expo, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post OpenAI spreads $600B cloud AI bet across AWS, Oracle, Microsoft appeared first on AI News. View the full article

Among the explosion of AI systems, AI web browsers such as Fellou and Comet from Perplexity have begun to make appearances on the corporate desktop. Such applications are described as the next evolution of the humble browser, and come with AI features built in; they can read and summarise web pages – and, at their most advanced – act on web content autonomously. In theory, at least, the promise of an AI browser is that it will speed up digital workflows, undertake online research, and retrieve information from internal sources and the wider internet. However, security research teams are concluding that AI browsers introduce serious risks into the enterprise that simply can’t be ignored. The problem lies in the fact that AI browsers are highly vulnerable to indirect prompt injection attacks. These are where the model in the browser (or accessed via the browser) receives instructions hidden in specially-crafted websites. By embedding text into web pages or images in ways humans find difficult to discren, AI models can be fed instructions in the form of AI prompts, or amendments to prompts that are input by the user. The bottom line for IT departments and decision-makers is that AI browsers are not yet suitable for use in the enterprise, and represent a significant security threat. Automation meets exposure In tests, researchers discovered that embedded text in online content is processed by the AI browser and is interpreted as instructions to the smart model. These instructions can be executed using the user’s privileges, so the greater the degree of access to information that the user has, the greater the risk to the organisation. The autonomy that AI gives users is the same mechanism that magnifies the attack surface, and the more autonomy, the greater the potential scope for data loss. For example, it’s possible to embed text commands into an image that, when displayed in the browser, could trigger an AI assistant to interact with sensitive assets, like corporate email, or online banking dashboards. Another test showed how an AI assistant’s prompt can be hijacked and made to perform unauthorised actions on the behalf of the user. These types of vulnerabilities clearly go against all principles of data governance, and are the most obvious example of how ‘shadow AI’ in the form of an unauthorised browser, poses a real threat to an organisation’s data. The AI model acts as a bridge between domains, and circumvents same-origin policies – the rule that prevents the access of data from one domain by another. Implementation and governance challenges The root of the problem is the merging of user queries in the browser with live data accessed on the web. If the LLM can’t distinguish between safe and malicious input, then it can blithely access data not requested by its human operator and act on it. When given agentic abilities, the consequences can be far-reaching, and could easily cause a cascade of malicious activity across the enterprise. For any organisation that relies on data segmentation and access control, a compromised AI layer in a user’s browser can circumvent firewalls, enact token exchanges, and use secure cookies in exactly the same way that a user might. Effectively, the AI browser becomes an insider threat, with access to all the data and facility of its human operator. The browser user will not necessarily be aware of activity ‘under the hood,’ so an infected browser may act for significant periods of time without detection. Threat mitigation The first generation of AI browsers should be regarded by IT teams in the same way they treat unauthorised installation of third-party software. While it is relatively easy to prevent specific software being installed by users, it’s worth noting that mainstream browsers such as Chrome and Edge are shipping with increased numbers of AI features in the form of Gemini (in Chrome) and Copilot (in Edge). The browser-producing companies are actively exploring AI-augmented browsing capabilities, and agentic features (that grant significant autonomy to the browser) will be quick to appear, driven by the need for competitive advantage between browser companies. Without proper oversight and controls, organisations are opening themselves to significant risk. Future generations of browsers should be checked for the following features: Prompt isolation, separating user intent from third-party web content before LLM prompt generation. Gated permissions. AI agents should not be able to execute autonomous actions, including navigation, data retrieval, or file access without explicit user confirmation. Sandboxing of sensitive browsing (like HR, finance, internal dashboards, etc.) so there is no AI activity in these sensitive areas. Governance integration. Browser-based AI has to align with data security policies, and the software should provide records to make agentic actions traceable. To date, no browser vendor has presented a smart browser with the ability to distinguish between user-driven intent, and model-interpreted commands. Without this, browsers may be coerced to act against the organisation by the use of relatively trivial prompt injection. Decision-maker takeaway Agentic AI browsers are presented as the next logical evolution in web browsing and automation in the workplace. They are designed deliberately to blur the distinction between user/human activity and become part of interactions with the enterprise’s digital assets. Given the ease with which the LLMs in AI browsers are circumvented and corrupted, the current generation of AI browsers can be regarded as dormant malware. The major browser vendors look set to embed AI (with or without agentic abilities) into future generations of their platforms, so careful monitoring of each release should be undertaken to ensure security oversight. (Image source: “Unexploded bomb!” by hugh llewelyn is licensed under CC BY-SA 2.0.) 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 AI browsers are a significant security threat appeared first on AI News. View the full article

For many *** executives, AI investment has become a necessity, not an experiment in innovation. Boards now demand evidence of measurable impact – whether through efficiency gains, revenue growth, or reduced operational risk. Yet, as Pete Smyth, CEO of Leading Resolutions notes, many SMEs treat AI as an exploratory exercise, not a structured business strategy. The result is wasted investment and a lack of demonstrable return. Business impact Enterprises implementing AI effectively are doing so with a focus on business outcomes. Instead of isolated pilots, they align initiatives with strategic goals – optimising operations and enhancing customer experience, for example. Leaders of organisations of any size can transform AI from a speculative technology into performance improvement by translating their ambitions into quantifiable metrics. Smyth gives examples that include automating routine analysis to reduce manual workflows, applying predictive analytics for inventory optimisation, or using natural language models to streamline customer service. The impact is measurable, he says: improved margins, faster decisions, and business resilience. Pete Smyth, Leading Resolutions Implementation & challenges According to Smyth’s Leading Resolutions, implementation success depends on priorities. The process begins with stakeholder engagement that identifies potential uses for AI in different departments. Each idea is evaluated for business value and readiness to implement; these processes produce a shortlist for potential pilot schemes. Next comes structured value assessment, combining cost-benefit analysis with execution feasibility and risk tolerance. Leaders should agree on the metrics that would define success before any pilot begins. These might include tracking KPIs (cost reduction, customer retention, productivity gains, etc.). Once validated, AI’s use can be scaled carefully in discrete business units. Strategic takeaway For data leaders and business decision-makers, measurable ROI requires a practically-based shift from experimentation to operational accountability. Focus should be on three principles, Smyth posits: Tie AI projects directly to business outcomes with pre-agreed KPIs. Embed governance, risk controls, and explainability early. Build an AI culture grounded in data quality, collaboration, and evidence-based decision-making. As enterprises navigate tighter regulation and rising AI expectations, success depends not on how much they invest, but how effectively they quantify and scale positive results. Moving from speculative ambition to measurable performance is the hallmark of credible AI implementation. (Main image source: “M4 AT Night” by Paulio Geordio is licensed under CC BY 2.0.) 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 From ambition to accountability: Quantifying AI ROI in strategy appeared first on AI News. View the full article

AI’s effects on continuous development and deployment pipelines are becoming difficult to ignore. However, decision-makers in software development functions need to consider a broad range of elements when considering the uses of the technology. The challenges of deploying AI at scale Deploying artificial intelligence isn’t the same as deploying, for example, a web app. Traditional software updates are usually deterministic: once code passes tests, everything works as it’s meant to. With AI and machine learning, outputs can vary because models depend on ever-changing data and complex statistical behaviour. Some unique challenges you’ll face include: Data drift: Your training data may not match real-world use, causing performance to decline. Model versioning: Unlike simple code updates, you need to track both the model and the data it was trained on. Long training times: Iterating on a new model can take hours or even days, slowing down releases. Hardware needs: Training and inference often require GPUs or specialised infrastructure. Monitoring complexity: Tracking performance in production means watching not just uptime but also accuracy, bias, and fairness. The challenges mean you can’t treat AI like traditional software. You need machine learning pipelines built with automation and monitoring. Applying DevOps principles to AI systems DevOps was designed to bring developers and operations closer by promoting automation, collaboration, and fast feedback loops. When you bring these principles to AI, so AI and DevOps, you create a foundation for scalable machine learning deployment pipelines. Some DevOps best practices translate directly: Automation: Automating training, testing, and deployment reduces manual errors and saves time. Continuous integration: Code, data, and model updates should all be integrated and tested regularly. Monitoring and observability: Just like server uptime, models need monitoring for drift and accuracy. Collaboration: Data scientists, engineers, and operations teams need to work together in the same cycle. The main difference between DevOps and MLOps lies in the focus. While DevOps centres on code, MLOps is about managing models and datasets alongside code. MLOps extends DevOps to address challenges specific to machine learning pipelines, like data validation, experiment tracking, and retraining strategies. Designing a continuous deployment pipeline for machine learning When building a continuous deployment system for ML, you need to think beyond just code. Gone are the days of just needing to know how to programme and code; now it’s about much more. Having an artificial intelligence development company that can implement these stages for you is crucial. A step-by-step framework could look like this: Data ingestion and validation: Collect data from multiple sources, validate it for quality, and ensure privacy compliance. For example, a healthcare company might verify that patient data is anonymised before use. Model training and versioning: Train models in controlled environments and store them with a clear version history. Fintech companies often keep a strict record of which datasets and algorithms power models that impact credit scoring. Automated testing: Validate accuracy, bias, and performance before models move forward. This prevents unreliable models from reaching production. Deployment to staging: Push models to a staging environment first to test integration with real services. Production deployment: Deploy with automation, often using containers and orchestration systems like Kubernetes. Monitoring and feedback loops: Track performance in production, watch for drift, and trigger retraining when thresholds are met. By designing an ML pipeline this way, you minimise risks, comply with regulations, and ensure reliable performance in high-stakes industries like healthcare and finance. The Role of a dedicated development team in MLOps You may wonder whether you need a dedicated software development team for MLOps or if hiring consultants is enough. The reality is that one-off consultants often provide short-term fixes, but machine learning pipelines require ongoing attention. Models degrade over time, new data becomes available, and deployment environments evolve. A dedicated team provides long-term ownership, cross-functional expertise, faster iteration, and risk management. Having a dedicated software development team that knows what it’s doing, how it’s doing it, and can keep doing it for you in the long run is ideal and works a lot better than having one-off consultants. Best practices for successful DevOps in AI Even with the right tools and teams, success in DevOps for AI depends on following solid best practices. These include: Version everything: Code, data, and models should all have clear version control. Test for more than accuracy: Include checks for fairness, bias, and explainability. Use containers for consistency: Containerising ML pipelines ensures models run the same in every environment. Automate retraining triggers: Set thresholds for data drift or performance declines that trigger retraining jobs automatically. Integrate monitoring into pipelines: Collect metrics on latency, accuracy, and use in real time. Collaborate in roles: Encourage shared responsibility between data scientists, engineers, and operations teams. Plan for scalability: Build pipelines that can handle growing datasets and user demand without major rework. These practices transform a machine learning pipeline from experimental systems into production-ready infrastructure. Conclusion The future of artificial intelligence depends on a reliable and scalable machine learning deployment pipeline. As a business, it’s paramount to implement AI in highly-specific ways to create digital services and products. The post DevOps for AI: Continuous deployment pipelines for machine learning systems appeared first on AI News. View the full article

At the APEC CEO Summit, NVIDIA said it is working with public agencies and private companies to build sovereign AI infrastructure across South Korea. The plan includes hundreds of thousands of NVIDIA GPUs across sovereign clouds and AI factories for areas like automotive, manufacturing and telecommunications. “Korea’s leadership in technology and manufacturing positions it at the heart of the AI industrial revolution — where accelerated computing infrastructure becomes as vital as power grids and broadband,” said Jensen Huang, founder and CEO of NVIDIA. “Just as Korea’s physical factories have inspired the world with sophisticated ships, cars, chips and electronics, the nation can now produce intelligence as a new export that will drive global transformation.” “Now that AI has gone beyond mere innovation and become the foundation of future industries, South Korea stands at the threshold of transformation,” said Bae Kyung-hoon, Korea Deputy Prime Minister, and Minister of Science and Information and Communication Technologies. The government plans to deploy up to 50,000 new NVIDIA GPUs to support sovereign AI programs for businesses and research groups. The first phase includes 13,000 NVIDIA Blackwell and other GPUs through providers such as NAVER Cloud, NHN Cloud and Kakao. The expansion includes efforts to build a National AI Computing Center. Startups, researchers and other organisations will be able to use this sovereign infrastructure to train models and build new applications. NVIDIA is also working with Samsung, SK Telecom, ETRI, KT, LGU+ and Yonsei University on AI-RAN and 6G network research. The work focuses on shifting some computing tasks from devices to network base stations, which may reduce battery drain and lower computing costs across sovereign AI services. Major companies build sovereign AI factories Large corporations in Korea are investing in advanced AI infrastructure for chip production, network operations and digital manufacturing tools that support the country’s sovereign computing goals. NVIDIA and Samsung plan to build a new AI factory that connects chip manufacturing with accelerated computing. The system will run more than 50,000 NVIDIA GPUs and support data-driven production methods, including predictive maintenance and process improvements across chip fabs. “We are at the dawn of the AI industrial revolution — a new era that will redefine how the world designs, builds and manufactures,” said Jensen Huang. Jay Y. Lee, executive chairman of Samsung Electronics, added, “From Samsung’s DRAM for NVIDIA’s game-changing graphics card in 1995 to our new AI factory, we are thrilled to continue our longstanding journey with NVIDIA in leading this transformation.” Samsung plans to use NVIDIA CUDA-X libraries, along with software from Synopsys, Cadence and Siemens, to speed circuit design and manufacturing workflows. It will also use NVIDIA Omniverse to create digital twins of factories and equipment for real-time simulation, testing and logistics planning — all supporting wider sovereign AI adoption. NVIDIA’s cuLitho library is being integrated into Samsung’s computational lithography tools. The collaboration has led to major gains in performance, supporting faster scaling in chip production. Samsung is also developing large language models that run across hundreds of millions of Samsung devices, supporting translation and other reasoning tasks. The company plans to expand into robotics using NVIDIA Isaac Sim, NVIDIA Cosmos and the Jetson Thor edge platform, which may strengthen its position in sovereign AI systems. SK Group expands AI capacity SK Group is building an AI factory that will include more than 50,000 NVIDIA GPUs, with completion expected by late 2027. The facility will support SK subsidiaries and outside clients through GPU-as-a-service offerings that align with South Korea’s sovereign AI strategy. NVIDIA and SK are also working together on next-generation high-bandwidth memory for GPUs. “SK Group is working with NVIDIA to make AI the engine of a profound transformation that will enable industries across Korea to transcend traditional limits of scale, speed and precision,” said Chey Tae-Won, chairman of SK Group. SK Telecom plans to build an industrial AI cloud using NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. The platform will support semiconductor manufacturing, digital twins and internal AI agents. SK hynix is using NVIDIA PhysicsNeMo tools to support chip design simulations, aiming to improve accuracy and speed. It is also testing NVIDIA Blackwell GPUs with Synopsys software and building autonomous fab digital twins. To support workers, SKT is developing a foundation model called A.X., built with NVIDIA NIM microservices and NVIDIA AI Enterprise. The model will power internal agents to assist thousands of employees across chip development and operations. Hyundai Motor Group plans new AI factory NVIDIA and Hyundai Motor Group are expanding their partnership to support autonomous driving, factory automation and robotics. Hyundai plans to build an AI factory using NVIDIA Blackwell GPUs for integrated training, simulation and deployment. “AI is revolutionising every facet of every industry, and in transportation alone — from vehicle design and manufacturing to robotics and autonomous driving — NVIDIA’s AI and computing platforms are transforming how the world moves,” said Jensen Huang. The companies expect joint investment of about $3 billion to grow national physical AI capabilities. The plan includes an NVIDIA AI Technology Center, Hyundai’s Physical AI Application Center and new data centres. These programs aim to help train a new generation of AI talent. Hyundai will use NVIDIA Omniverse Enterprise to build digital twins of factories, supporting virtual testing, robot integration and predictive maintenance. It will also use NVIDIA DRIVE AGX Thor for in-vehicle AI systems, including driver assistance and infotainment features. Growth of sovereign AI models NAVER Cloud plans to deploy more than 60,000 GPUs for sovereign and physical AI work. The company will build industry-targeted models for areas such as shipbuilding and public safety. The Ministry of Science and ICT is also leading a Sovereign AI Foundation Models project using NVIDIA NeMo and open Nemotron datasets. Partners include LG AI Research, NC AI, SK Telecom and Upstage. These models will support language and reasoning tasks. LG is working with NVIDIA on physical AI research and will support startups and researchers using its EXAONE models, including healthcare applications. Quantum and scientific research KISTI plans to use NVIDIA accelerated computing in its sixth national supercomputer, HANGANG. The institute will support NVQLink, an open architecture for connecting quantum processors with GPU clusters. It will also develop scientific foundation models and explore physics-informed AI tools using NVIDIA PhysicsNeMo. NVIDIA and local partners are forming a startup alliance through the NVIDIA Inception program. Members will gain access to accelerated computing resources from cloud partners like SK Telecom, along with support from venture firms. NVIDIA also plans to take part in the N-Up AI startup incubation program from the Ministry of SMEs and Startups. (Photo by Nvidia) See also: Migrating AI from Nvidia to Huawei: Opportunities and trade-offs 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 NVIDIA and South Korea align on sovereign AI at APEC CEO Summit appeared first on AI News. View the full article

The AI chip wars just got a new heavyweight contender. Qualcomm, the company that powers billions of smartphones worldwide, has made an audacious leap into AI data centre chips – a market where Nvidia has been minting money at an almost unfathomable rate and where fortunes rise and fall on promises of computational supremacy. On October 28, 2025, Qualcomm threw down the gauntlet with its AI200 and AI250 solutions, rack-scale systems designed specifically for AI inference workloads. Wall Street’s reaction was immediate: Qualcomm’s stock price jumped approximately 11% as investors bet that even a modest slice of the exploding AI infrastructure market could transform the company’s trajectory. The product launch could redefine Qualcomm’s identity. The San Diego chip giant has been synonymous with mobile technology, riding the smartphone wave to dominance. But with that market stagnating, CEO Cristiano Amon is placing a calculated wager on AI data centre chips, backed by a multi-billion-dollar partnership with a Saudi AI powerhouse that signals serious intent. Two chips, two different bets on the future Here’s where Qualcomm’s strategy gets interesting. Rather than releasing a single product and hoping for the best, the company is hedging its bets with two distinct AI data centre chip architectures, each targeting different market needs and timelines. The AI200, arriving in 2026, takes the pragmatic approach. Think of it as Qualcomm’s foot in the door – a rack-scale system packing 768 GB of LPDDR memory per card. That massive memory capacity is crucial for running today’s memory-hungry large language models and multimodal AI applications, and Qualcomm is betting that its lower-cost memory approach can undercut competitors on total cost of ownership while still delivering the performance enterprises demand. But the AI250, slated for 2027, is where Qualcomm’s engineers have really been dreaming big. The solution introduces a near-memory computing architecture that promises to shatter conventional limitations with more than 10x higher effective memory bandwidth. For AI data centre chips, memory bandwidth is often the bottleneck that determines whether your chatbot responds instantly or leaves users waiting. Qualcomm’s innovation here could be a genuine game-changer – assuming it can deliver on the promise. “With Qualcomm AI200 and AI250, we’re redefining what’s possible for rack-scale AI inference,” said Durga Malladi, SVP and GM of technology planning, edge solutions & data centre at Qualcomm Technologies. “The innovative new AI infrastructure solutions empower customers to deploy AI at unprecedented TCO, while maintaining the flexibility and security modern data centres demand.” The real battle: Economics, not just performance In the AI infrastructure arms race, raw performance specs only tell half the story. The real war is fought on spreadsheets, where data centre operators calculate power bills, cooling costs, and hardware depreciation. Qualcomm knows this, and that’s why both AI data centre chip solutions obsess over total cost of ownership. Each rack consumes 160 kW of power and employs direct liquid cooling – a necessity when you’re pushing this much computational power through silicon. The systems use PCIe for internal scaling and Ethernet for connecting multiple racks, providing deployment flexibility whether you’re running a modest AI service or building the next ChatGPT competitor. Security hasn’t been an afterthought either; confidential computing capabilities are baked in, addressing the growing enterprise demand for protecting proprietary AI models and sensitive data. The Saudi connection: A billion-dollar validation Partnership announcements in tech can be vapour-thin, but Qualcomm’s deal with Humain carries some weight. The Saudi state-backed AI company has committed to deploying 200 megawatts of Qualcomm AI data centre chips – a figure that analyst Stacy Rasgon of Sanford C. Bernstein estimates translates to roughly $2 billion in revenue for Qualcomm. Is $2 billion transformative? In the context of AMD’s $10 billion Humain deal announced the same year, it might seem modest. But for a company trying to prove it belongs in the AI infrastructure conversation, securing a major deployment commitment before your first product even ships is validation that money can’t buy. “Together with Humain, we are laying the groundwork for transformative AI-driven innovation that will empower enterprises, government organisations and communities in the region and globally,” Amon declared in a statement that positions Qualcomm not just as a chip supplier, but as a strategic technology partner for emerging AI economies. The collaboration, first announced in May 2025, transforms Qualcomm into a key infrastructure provider for Humain’s ambitious AI inferencing services – a role that could establish crucial reference designs and deployment patterns for future customers. Software stack and developer experience Beyond hardware specifications, Qualcomm is betting on developer-friendly software to accelerate adoption. The company’s AI software stack supports leading machine learning frameworks and promises “one-click deployment” of models from Hugging Face, a popular AI model repository. The Qualcomm AI Inference Suite and Efficient Transformers Library aim to remove integration friction that has historically slowed enterprise AI deployments. David vs. Goliath (and another Goliath?) Let’s be honest about what Qualcomm is up against. Nvidia’s market capitalisation has soared past $4.5 trillion, a valuation that reflects years of AI dominance and an ecosystem so entrenched that many developers can’t imagine building on anything else. AMD, once the scrappy challenger, has seen its shares more than double in value in 2025 as it successfully carved out its own piece of the AI pie. Qualcomm’s late arrival to the AI data centre chips party means fighting an uphill battle against competitors who have battle-tested products, mature software stacks, and customers already running production workloads at scale. The company’s smartphone focus, once its greatest strength, now looks like strategic tunnel vision that caused it to miss the initial AI infrastructure *****. Yet market analysts aren’t writing Qualcomm’s obituary. Timothy Arcuri of UBS captured the prevailing sentiment on a conference call: “The tide is rising so fast, and it will continue to rise so fast, it will lift all boats.” Translation: the AI market is expanding so rapidly that there’s room for multiple winners – even latecomers with compelling technology and competitive pricing. Qualcomm is playing the long game, betting that sustained innovation in AI data centre chips can gradually win over customers looking for alternatives to the Nvidia-AMD duopoly. For enterprises evaluating AI infrastructure options, Qualcomm’s emphasis on inference optimisation, energy efficiency, and TCO presents an alternative worth watching – particularly as the AI200 approaches its 2026 launch date. (Photo by Qualcomm) See also: Migrating AI from Nvidia to Huawei: Opportunities and trade-offs 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 Qualcomm unveils AI data centre chips to crack the Inference market appeared first on AI News. View the full article

It’s no longer news that AI is transforming how people communicate at work. The bad (and less common) news, however, is that AI is also making those conversations harder to control. From chat apps to collaboration tools, employees exchange thousands of messages every day, many of which now pass through AI systems that summarise, analyse, or even respond on their behalf. For enterprises, that creates a new kind of exposure: Communication data that is intelligent, unstructured, and often ungoverned. Dima Gutzeit, CEO of business communications platform provider LeapXpert, believes the future of enterprise communication depends on solving this challenge. “AI has made conversation the most valuable dataset inside organisations,” he said. “But without structure and governance, that value quickly turns into risk.” The enterprise communication blind spot For years, corporate communications were treated as either static records – emails stored in archives – or ephemeral exchanges that disappeared after use. The rise of AI has changed that. Tools like Microsoft’s Copilot and Zoom’s AI Companion now interpret tone, context, and intent in real-time, turning chat history into searchable knowledge. But for many companies, that same intelligence is emerging in silos, without visibility or control. “Every enterprise is adopting AI somewhere in its communication stack,” Gutzeit said. “The problem is that few have a unified way to manage it in all channels, especially when client conversations happen on platforms like WhatsApp or iMessage.” That lack of oversight has real-world consequences with far-reaching impact. According to a 2025 Kiteworks survey, 83% of organisations admit they have limited visibility into how employees use AI tools at work, and nearly half have already experienced at least one AI-related data incident. The challenge here isn’t just data loss, but also accountability. Turning conversations into intelligence LeapXpert’s platform aims to close that gap through what the company calls “Communication Data Intelligence”. The system captures and consolidates all external client communications, whether from WhatsApp, WeChat, iMessage, or Microsoft Teams, into a single, governed environment. In this framework, LeapXpert’s proprietary AI engine, Maxen analyses messages for sentiment, intent, and compliance signals, and maintains full auditability. That means that every conversation can be understood responsibly. Relationship managers, compliance officers, and legal teams can see the same transparent record of who said what, when, and why. The AI can also detect anomalies, flag potential policy violations, and generate summaries for faster reviews. “Think of it as bringing context to compliance,” Gutzeit said. “Our goal is not to replace human communication, but to make it smarter, safer, and accountable.” Results in the real world LeapXpert backs its claims about its communications data intelligence concept with customer proofs from the field. In one such case, a North American investment management firm operating under SEC and FINRA oversight implemented the LeapXpert platform recently to consolidate its messaging systems. Before deployment, the compliance team manually sampled conversations from several archives – a process that consumed hours each day. However, after integrating LeapXpert’s platform, all communications were consolidated into a single, auditable system, resulting in a 65% reduction in manual review time and an improvement in audit response times from days to hours. More importantly, the firm also gained real-time visibility into emerging conduct risks, while employees continued using the communications channels their clients preferred. Gutzeit said such results highlight the growing industry reality that regulated enterprises can no longer afford to separate innovation from compliance. “They have to move together,” he said. Governing the AI era of communication The rise of embedded AI features in everyday tools adds another layer of urgency. Platforms like Slack, Salesforce, and Microsoft Teams now include generative assistants that summarise messages or recommend actions; functions that may automatically process sensitive data. Without clear governance, these features can introduce the same risks that external tools once did. That is where Gutzeit says LeapXpert’s architecture stands apart. The platform operates on a zero-trust framework, encrypting every message in transit and at rest. Customers retain full data ownership through bring-your-own-key encryption, while AI operations run in secure, isolated environments. “Our systems are built so enterprises can benefit from AI without surrendering control of their data,” Gutzeit said. The path ahead As AI continues to permeate enterprise communication, Gutzeit expects governance to evolve from a defensive measure to a source of business intelligence. “We’re entering a phase where AI will understand communication, not just record it,” he said. “That means compliance officers and business leaders can both derive value from the same dataset.” Gutzeit also sees it as the next logical step in the evolution of enterprise communication. “AI will only be transformative if it’s trusted,” he noted. “Transparency, auditability, and context are what make that possible.” For enterprises navigating the tension between innovation and oversight, LeapXpert offers the proposition of AI that listens, understands, and stays accountable. Image source: Unsplash The post How LeapXpert uses AI to bring order and oversight to business messaging appeared first on AI News. View the full article

For all the progress in artificial intelligence, most video security systems still fail at recognising context in real-world conditions. The majority of cameras can capture real-time footage, but struggle to interpret it. This is a problem turning into a growing concern for smart city designers, manufacturers and schools, each of which may depend on AI to keep people and property safe. Lumana, an AI video surveillance company, believes the fault in these systems lies deep in the foundations of how they are built. “Traditional video platforms were created decades ago to record footage, not interpret it,” said Jordan Shou, Lumana’s Vice President of Marketing. “Adding AI on top of outdated infrastructure is like putting a smart chip in a rotary phone. It might function, but it will never be truly intelligent or reliable enough to understand what’s being captured or help teams make smarter real-time decisions.” Big consequences When traditional video security systems layer AI on older infrastructure, false alerts and performance issues arise. Alerts and missed detections are not just technical hiccups, but risks that can have devastating consequences. Shou points to a recent case where a school surveillance system, which used an AI add-on for gun detection, misidentified a harmless object for a weapon, setting off an unnecessary police response. “Every mistake, whether it’s a missed event or a false alert, which leads to improper response, erodes trust,” he said. “It wastes time, money, and can traumatise people who did nothing wrong.” Errors can also be costly. Each false alarm forces teams to pause real work and investigate, a process that can drain millions from public safety and operational budgets every year. Building a smarter foundation Instead of layering AI on top of old video security frameworks, Lumana rebuilt the infrastructure itself with an all-in-one platform that combines modern video security hardware, software, and proprietary AI. The company’s hybrid-cloud design connects any security camera to GPU-powered processors and adaptive AI models that operate at the edge – meaning they are located as close as possible to where the footage is captured. The result, Shou says, is faster performance and more accurate analysis. Each camera becomes a continuous-learning device that improves over time, understanding motion, behaviour, and patterns unique to its environment. “The issue is that most of today’s video surveillance systems use static, off-the-shelf AI models that were only designed to work in specific environments. AI shouldn’t need a perfect lab environment to work,” Shou explained. “It should work in real-world conditions and adapt based on the video data that’s coming in. That’s why, when customers compare Lumana to their existing or other AI systems, the difference and performance gaps are immediately clear.” The company’s design also prioritises privacy. All data is encrypted, governed by access controls, and compliant with SOC 2, HIPAA, and NDAA standards. Customers can disable facial or biometric tracking if they choose. “Our focus is on actions, not identities,” Shou said. Real-world use cases Lumana’s systems have been deployed in several industries. One of its most visible projects is with JKK Pack, a 24-hour packaging manufacturer that uses security cameras to monitor safety and operational efficiency in its facilities. Before Lumana’s deployment, cameras only recorded incidents for later review, which led to missed events and reactive incident response. After the upgrade, the same hardware could detect unsafe movements, equipment faults, or manufacturing bottlenecks in real-time. The company reported 90% faster investigations and alerts delivered in under a second which dramatically improved response to safety incidents, without replacing a single camera. In another deployment, a grocery retailer integrated Lumana’s AI into its existing camera network to flag unusual point-of-***** activity, like repeat voids, and to correlate those events with visual evidence. The system reduced shrinkage and improved employee accountability by providing real-world examples of policy violations. Beyond manufacturing, Lumana’s system has been used at large public events, in restaurants, and for municipal operations. In cities, it helps identify ******** dumping and fires; in quick-service chains, it monitors kitchen safety and food handling. A broader push for reliable AI video security Lumana’s work comes at a time when accuracy and accountability are replacing speed as the top priorities for enterprise AI. A recent study from F5 found that only 2% of companies consider themselves fully ready to scale AI, with governance and data security cited as the main challenges. That caution is reflected in the market, with analysts warning that as AI takes on more decision-making, systems must remain “auditable, transparent, and free from bias.” Lumana’s architecture echoes the call for accountability, blending performance and control with data governance and cybersecurity in an easy-to-deploy solution that enhances existing security camera infrastructure, helping organisations extract immediate value from AI video. The next step in machine vision Shou said Lumana’s next stage of development aims to move from detection and understanding to predicting. “The next evolution of AI video will be about reasoning,” he said. “The ability to grasp context in real time, provide actionable and impactful insights from the video data collected, will change how we think about safety, operations, and awareness.” For Lumana, the goal is not just teaching AI how to see better, but to help it understand what it is seeing and letting those who rely on that video data to make smarter, faster decisions. Image source: Unsplash The post How Lumana is redefining AI’s role in video surveillance appeared first on AI News. View the full article

The acquisition of a legacy platform like AOL by Bending Spoons shows the latent value of long-standing digital ecosystems. AOL’s 30 million monthly active users represent an enduring brand and a data-rich resource that can be used in AI-driven services. That statement is true only if the data is properly governed and integrated. Such deals may blend nostalgia with business advantage, but present new compliance and cybersecurity risks that enterprises need to address. By acquiring AOL from Yahoo, Bending Spoons moves to consolidate high-retention consumer technologies in its expanding digital portfolio. As companies turn increasingly to synthetic data to feed their AI’s learning corpus, the deal shows a different tactic, one of using established data assets and user bases to accelerate AI personalisation, advertising efficiency, and digital identity information gathering. It illustrates how older platforms – perhaps written off as legacy – can become profitable fuel for innovation when combined with cloud-native architectures and machine learning models. Bending Spoons has financed its expansion strategy with a $2.8 billion debt package from major global banks that include J.P. Morgan, BNP Paribas, and HSBC. There’s clearly growing lender confidence in the long-term monetisation of data, unlike during the ‘dot.com’ ***** and bust, where the emphasis and interest was in purely software products. The acquisition, expected to close by year-end, follows Bending Spoons’ planned purchase of Vimeo. The two deals, if they go through, position the company as a major consolidator of internet assets. Implementation and operational challenges Integrating decades-old infrastructure like AOL’s presents technical challenges. Data migration from legacy email systems in line with current-day security protocols and compliance requirements needs careful stewardship. There’s also the not-insignificant issue of retraining staff for AI data stewardship on data that comes with significant buy-in from trusting service users. As with any digital acquisition, therefore, Bending Spoons’ success will depend on managing the technical and cultural dimensions of integration. Without strong governance, promising legacy platforms risk becoming compliance liabilities. Early in any acquisition cycle, there will have been preparatory work in mapping data lineage, running integration and interoperability audits, and significant governance discussions. It’s worth noting that many integration pilots stall without shared accountability between technology and business functions: It’s easier to covet data than to work out how it can be put to business use, especially when the best an acquirer can hope for are limited examples of what they might get, once the ink has dried on the cheque. Vendor and ecosystem context Although Bending Spoons operates independently of major enterprise AI ecosystems, the logic of its acquisition aligns with Microsoft’s integration of LinkedIn data into Azure AI Foundry, and IBM’s efforts to reinvigorate legacy data with watsonx. AOL’s customer base and behavioural data could feasibly hold value with cloud analytics, customer profiling, and identity management frameworks, on even off-the-shelf platforms like AWS Bedrock, Azure, or Google Vertex AI. Executive takeaway Legacy platforms are not obsolete but they are often underused and undervalued. The differentiator lies in how organisations integrate historical data into modern AI governance and value delivery. Executives may see the AOL acquisition as a nostalgia play, but it’s a more hard-nosed imagining of a pure data asset. Perhaps the next wave of competitive advantage may come not from building new systems, but from reinterpreting older software and information that’s sometimes disregarded, simply because it’s not the latest-and-greatest ‘thing.’ (Image source: “Spoon” by felixtsao is licensed under CC BY 2.0.) 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 Bending Spoons’ acquisition of AOL shows the value of legacy platforms appeared first on AI News. View the full article

Citizens of Thailand can now access the Sora app, giving local creators an early look at OpenAI’s new AI video tool in Asia. Thailand already has an active creative scene, and this launch is meant to support more visual storytelling from the region. The app’s rollout also includes Vietnam and Taiwan. Sora first arrived in the US and Canada in early September, and many users there have already shared clips. The app has since passed one million downloads in under five days, according to a social media post by Sora head Bill Peebles, who noted that it reached that milestone even faster than ChatGPT did at launch, despite requiring users to be invited to use the app at launch. People in Thailand can download the app for free on iOS with no invite code. For now, use limits are relatively generous, though those limits may change. The app is powered by Sora 2, a video generation model that can produce ‘original’ clips, remix existing creations, and suggest content through a personal feed. Users can also appear directly inside scenes through a feature called Cameos, which requires a one-time check to confirm identity and likeness. The app supports Thai language input. Cameos have quickly become a popular feature among early testers as they offer a playful way to interact and connect with friends. Thai creator Woody Milintachinda said, “Sora allows me to bring ideas to life in a way that immediately resonates with audiences. They can see and feel the story unfold. It has also given me a unique platform to share my experiences with a wide community of creators and storytellers not just in Thailand but the world, inspiring new forms of connection and creativity. With Sora, the creative possibilities truly feel limitless.” To go with this release, the app now includes Character Cameos, with which users can turn nearly anything into a reusable character, such as a ****, drawing, personal item, or original design created inside Sora. After uploading a video of the character, users can assign permissions that are separate from their personal likeness. That character can stay private, be shared only with followers, or be opened to everyone on the platform. Once named, the character can appear in any future video. To mark the Halloween season, the app launches with a starter pack that includes classic characters like Dracula, Frankenstein’s monster, Ghost, Witch, and Jack-O-Lantern. The company says it plans to bring Sora to Thailand with responsibility in mind. The feed is designed to encourage creation rather than passive viewing, aimed at accounts users follow. The aim is not to increase screen time but to spark creative output, the company states. Users can keep control of their likeness when using Cameos, deciding who can use it, and the account holder can remove access or take down any video that includes their likeness at any time. Videos made with a cameo of the user created by someone else remain visible to the user. Videos produced in Sora include a visible, animated watermark and an invisible C2PA digital watermark. The hidden version cannot be added to content that was not created in Sora, helping confirm which clips were created on the platform. For teens, the app applies daily limits on how many generated videos appear in their feed. Cameos also come with stricter rules for this demographic. Safety systems exist, and human moderators can review bullying cases. Parents can use ChatGPT-based controls to adjust feed limits, turn off personalisation, and manage direct message settings. (Photo by Mariia Shalabaieva) See also: OpenAI unveils open-weight AI safety models for developers 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 Thailand becomes one of the first in Asia to get the Sora app appeared first on AI News. View the full article

Samsung’s semiconductor recovery has materialised during the third quarter of 2025, with the South Korean tech giant posting an operating profit of KRW 12.2 trillion (US$8.6 billion) – more than double the previous quarter and ending a streak of four consecutive quarterly declines in its chip division. The turnaround centred on Samsung’s Device Solutions division, which reported KRW 33.1 trillion in revenue and KRW 7.0 trillion in operating profit, an over tenfold increase from the June quarter. The Memory Business achieved what Samsung described as “record-high quarterly revenue,” driven by expanded sales of high-bandwidth memory (HBM3E) chips and server solid-state drives – both important components for artificial intelligence infrastructure. But this wasn’t simply a story of rising tides lifting all boats. Samsung’s semiconductor recovery reflects calculated strategic pivots made during its downturn, market dynamics that finally shifted in its favour, and intense competitive pressures that forced the company to accelerate its AI chip roadmap. The road back from the slump Samsung’s journey to this quarter’s performance began in a different place. Throughout 2024 and into early 2025, the company faced multiple headwinds: a brutal memory chip glut that collapsed prices, delayed qualification of its HBM products with key customers, and the company seeing rival SK Hynix capture early leadership in AI memory chips. The low point came in the second quarter of 2025, when Samsung’s chip division reported operating profit that had analysts questioning whether the company had lost its technological edge. SK Hynix had seized the top spot in the memory market for the first time, fueled by its early success supplying HBM chips to Nvidia’s AI accelerators. MS Hwang, research director at Counterpoint Research, contextualised Samsung’s third-quarter performance as “a clear result of a broader memory market ***** and rising prices for general-purpose memory.” But Hwang’s firm also noted that Samsung had reclaimed the top spot in the memory market from SK Hynix during Q3, suggesting the semiconductor recovery involved more than just favourable market conditions. HBM: From laggard to mass production Samsung’s ability to reverse its HBM fortunes proved central to the turnaround. The company confirmed that HBM3E is now “in mass production and being sold to all related customers,” while HBM4 samples are “simultaneously being shipped to key clients.” Reports emerged in late September that Samsung had passed Nvidia’s qualification tests for advanced high-bandwidth memory chips – a important milestone that had eluded the company for months. While Samsung hasn’t confirmed the Nvidia qualification publicly, the timing aligns with the acceleration in HBM sales reflected in Q3 results. During the company’s earnings call, a Samsung executive outlined the demand environment: “We expect data centre companies to continuously expand their hardware investment because of the ongoing competition to secure AI infrastructure. Therefore, our AI-related server demand keeps growing, and this demand significantly exceeds industry supply.” That supply-demand imbalance has created pricing power that Samsung lacked during its declining quarters. The company specifically cited “a favourable price environment” and “notably reduced one-off costs like inventory value adjustments” as contributors to higher profits. Beyond memory: Foundry progress and challenges Samsung’s semiconductor recovery extended beyond memory chips. The Foundry Business, which manufactures chips designed by other companies, “posted a significant improvement in earnings in Q3 2025, stemming from a reduction in one-off costs and better fab use.” The division also achieved “record-high customer orders, mainly on advanced nodes.” The foundry business is ramping up mass production of 2-nanometer Gate-All-Around (GAA) products, an important technology that helps maintain competitiveness against TSMC, the company that dominates the foundry market. Samsung indicated it would begin operations at its fab in Taylor, Texas, “in a timely manner” in 2026. However, the System LSI Business, which designs Samsung’s Exynos processors and image sensors, saw earnings stall “due to seasonality and customer inventory adjustments.” What this means for 2026 Samsung’s guidance for the coming year reflects confidence that the semiconductor recovery has staying power. The Memory Business will “focus on the mass production of HBM4 products with differentiated performance” while aiming to “scale out the HBM sales base.” The company plans capacity expansion in its 1c manufacturing process to meet projected HBM4 demand increases. Consolidated revenue for the quarter reached KRW 86.1 trillion, a 15.4% increase from the previous quarter and 8.85% higher year-over-year. The Device eXperience division, which includes smartphones, contributed KRW 34.1 trillion in revenue, supported by the Galaxy Z Fold7 launch and strong flagship sales. Yet challenges persist. Samsung Display reported solid performance with KRW 8.1 trillion in revenue and KRW 1.2 trillion in operating profit, but the Visual Display business recorded an operating loss of KRW 0.1 trillion despite “solid sales growth of premium products,” citing “intensified competition.” The semiconductor recovery that Samsung achieved in Q3 2025 resolves the immediate crisis that threatened its market position. Whether the company can sustain this momentum while navigating intense competition from SK Hynix in HBM, TSMC in foundry, and emerging geopolitical pressures on the chip industry will determine if this quarter marked a true turning point or merely a reprieve. For now, Samsung has demonstrated that even after four quarters of decline, strategic execution and market timing can still produce a comeback. (Photo by Babak Habibi) See also: Samsung AI strategy delivers record revenue despite semiconductor headwinds 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 Inside Samsung’s semiconductor recovery: How AI demand reversed four quarters of decline appeared first on AI News. View the full article

AI is spreading through workplaces faster than any other technology in recent memory. Every day, employees connect AI technologies to enterprise systems, often without permission or oversight from IT security teams. The result is what experts call shadow AI – a growing web of tools and integrations that access company data unmonitored. Dr.Tal Shapira, Co founder and CTO at SaaS security and AI governance solution provider Reco, says this invisible sprawl could become one of the biggest threats facing organisations today, especially since the current speed of AI adoption has outpaced enterprise safeguards. “We went from ‘AI is coming’ to ‘AI is everywhere’ in about 18 months. The problem is that governance frameworks simply haven’t caught up,” Shapira said. The invisible risk inside company systems Shapira said most corporate security systems were designed for an older world where everything stayed behind firewalls and network borders. Shadow AI breaks that model because it works from the inside, hidden in the company’s own tools. Many modern AI tools connect straight into everyday SaaS platforms like Salesforce, Slack, or Google Workspace. While that is not a risk in itself, AI often does this through permissions and plug-ins that stay active after installation. Those ‘quiet’ links can keep giving AI systems access to company data, even after the person who set them up stops using them or leaves the organisation. That’s a big shadow AI problem. Shapira said: “The deeper issue is that these tools are embedding themselves into the company’s infrastructure, sometimes for months or years without detection.” The new class of risk is especially difficult to track as many AI systems are probabilistic. Instead of executing clear commands, AI makes predictions based on patterns, so their actions can change from one situation to the next, making them harder to review and control. When AI goes rogue The damage from shadow AI is already evident in real-world incidents. Reco recently worked with a Fortune 100 financial firm that believed its systems were secure and compliant. In days of deploying Reco’s monitoring, the company uncovered more than 1,000 unauthorised third-party integrations in its Salesforce and Microsoft 365 environments – over half of them powered by AI. One integration, a transcription tool connected to Zoom, had been recording every customer call, including pricing discussions and confidential feedback. “They were unknowingly training a third-party model on their most sensitive data,” Shapira noted. “There was no contract, no understanding of how that data was being stored or used.” In another case, an employee linked ChatGPT directly to Salesforce, allowing the AI to generate hundreds of internal reports in hours. That might sound efficient, but it also exposed customer information and sales forecasts to an external AI system. How Reco detects the undetected Reco’s platform is built to give companies full visibility into what AI tools are connected to their systems and what data those tools can access. It scans SaaS environments for OAuth grants, third-party apps, and browser extensions continuously. Once identified, Reco shows which users installed them, what permissions they hold, and whether the behaviour looks suspicious. If a connection appears risky, the system can alert administrators or revoke access automatically. “Speed matters because AI tools can extract massive amounts of data in hours, not days,” Shapira said. Unlike traditional security products that rely on network boundaries, Reco focuses on the identity and access layer. That makes it well suited for today’s cloud-first, SaaS-heavy organisations where most data lives outside the traditional firewall. A wider security wake-up call Industry analysts say Reco’s work reflects a larger trend in enterprise security: A shift from blocking AI to governing it. According to a recent Cisco report on AI readiness, in 2025 62% of organisations admitted they have little visibility into how employees are using AI tools at work, and nearly half have already experienced at least one AI-related data incident. As AI features become embedded in mainstream software – from Salesforce’s Einstein to Microsoft Copilot — the challenge grows. “You may think you’re using a trusted platform,” Shapira said, “but you might not realise that platform now includes AI features accessing your data automatically.” Reco’s system helps close the gap by monitoring sanctioned and unsanctioned AI activity, helping companies build a clearer picture of where their data is flowing, and why. Harnessing AI securely Shapira believes enterprises are entering what he calls the AI infrastructure phase – a ******* when every business tool will include some form of AI, whether visible or not. That makes continuous monitoring, least-privilege access, and short-lived permissions essential. “The companies that succeed won’t be the ones blocking AI,” he observed. “They’ll be the ones adopting it safely, with guardrails that protect both innovation and trust.” Shadow AI, he said, is not a sign of employee recklessness, but of how quickly technology has moved. “People are trying to be productive,” he said. “Our job is to make sure they can do that without putting the organisation at risk.” For enterprises trying to harness AI without losing control of their data, Reco’s message is simple: You can’t secure what you can’t see. Image source: Unsplash The post Reco wants to eliminate the blind spot of shadow AI appeared first on AI News. View the full article

Cursor has released its latest AI software development platform with a new multi-agent interface and the debut of its coding model, Composer. The new Composer model is described as a “frontier model”. Cursor claims it is four times faster than other models of similar intelligence. The company built it specifically for “low-latency agentic coding” within the Cursor environment. The company states that the model can complete most conversational turns in under 30 seconds. This speed is intended to improve the developer’s workflow. Early testers reported that the ability to iterate quickly with the model was a key benefit. They also apparently grew to trust Composer for handling complex and multi-step coding tasks. To achieve this performance, Composer was trained with a suite of powerful tools. One of the key tools mentioned is “codebase-wide semantic search”. This training, Cursor says, makes Composer much better at understanding and working in large, complex codebases—a common challenge for many generative AI coding assistants. The second major update is the new user interface. Upon opening the new version, users will notice a “more focused” design. The entire AI-driven software development experience in Cursor has been rebuilt to be “centered around agents rather than files”. This change in focus is designed to allow developers to concentrate on their desired outcomes, while the AI agents manage the underlying details and code implementation. For developers who still need to work directly with the code, the new layout retains the ability to open files easily. Users can also revert to the “classic IDE” view if they prefer. A core feature of Cursor’s new platform is its ability to run many AI agents in parallel without them interfering with one another. This functionality is powered by technologies like “git worktrees or remote machines”. Cursor also noted an interesting emergent strategy from this parallel approach. They found that assigning the same problem to multiple different models and then selecting the best solution “greatly improves the final output”. This is particularly effective for more difficult or complex tasks. The company acknowledges that as AI agents take on more of the coding workload, new bottlenecks have emerged for developers. The two biggest new challenges are “reviewing code and testing the changes”. Cursor 2.0 includes new features designed to start solving both of these problems. The interface has been simplified to make it “much easier to quickly review the changes an agent has made”. This allows developers to dive deeper into the code only when necessary. Cursor 2.0 also introduces a “native browser tool” that enables the AI agent to test its own work automatically. The agent can then iterate on its solution, running tests and making adjustments until it produces the “correct final result”. This marks a step towards a more autonomous development process, where agents can not only write code but also validate it. See also: OpenAI unveils open-weight AI safety models for developers 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 Expo, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post Cursor 2.0 pivots to multi-agent AI coding, debuts Composer model appeared first on AI News. View the full article

For many years, Nvidia has been the de facto leader in AI model training and inference infrastructure, thanks to its mature GPU range, the CUDA software stack, and a huge developer community. Moving away from that base is therefore a strategic and tactical consideration. Huawei AI represents an alternative to Nvidia, with the ******** company signalling an increasingly aggressive move into AI hardware, chips, and systems. This presents decision-makers with opportunities. For example: The company has unveiled its SuperPod clusters that link thousands of Ascend NPUs, with claims that data links, for example, are “62× quicker”, and that the offering is more advanced than Nvidia’s next-gen alternative. Huawei’s strategy emphasises its inference advantages. In domestic or alternative markets where export control or supply-chain risk makes a single-vendor (Nvidia) strategy less robust, the ******** company’s portfolio is the logical choice. Any migration to a Huawei-centred pipeline isn’t, however a simple a plug-in replacement. It would entail a shift in developer ecosystem and possible regional re-alignment. Business advantages of moving to a Huawei AI-centred pipeline When contemplating the shift, several business advantages may drive a final decision. Relying on one major vendor (namely, Nvidia) can incur risks: pricing leverage, export controls, supply shortages, or a single point of failure in innovation. Adopting or migrating to Huawei has the potential to provide negotiation leverage, avoid vendor lock-in, and offer access to alternate supply chains. That’s especially relevant in areas where Nvidia faces export restrictions. If an organisation operates in a region where Huawei’s ecosystem is stronger (e.g., China, parts of Asia) or where domestic incentives favour local hardware, shifting to Huawei could align with corporate strategy. For instance, ByteDance has begun training a new model primarily on Huawei’s Ascend 910B chips with notable success. Huawei’s technology focuses on inference and large-scale deployments, and thus may be better suited to long-term use, rather than occasional use of large infrastructures for training, followed by less intensive inference. If an organisation’s workloads are inference-heavy, a Huawei stack may offer advantages in cost and power. Moving Huawei’s internal clusters (e.g., CloudMatrix) have shown competitive results in select benchmarks. Risks and trade-offs While migration offers potential gains, several challenges exist. Nvidia’s CUDA ecosystem remains unmatched for tooling and community support, with Nvidia established as the go-to solution for most companies and businesses. Migrating to Huawei’s Ascend chips and CANN software stack may require re-engineering workloads, retraining staff, and adjusting frameworks. Those are not considerations to be taken lightly. Additionally, Huawei hardware still lags Nvidia in high-end benchmarks. One ******** firm reportedly needed 200 engineers and six months to port a model from Nvidia to Huawei, yet only achieved about 90% of prior performance. The wholesale rebuilding of development pipelines will incur engineering and operational costs. If significant investment in Nvidia hardware and CUDA-optimised workflows exists, switching will not yield short-term savings. And while use of Huawei technologies mitigates dependency on Western chips, it may introduce other regulatory risks given the controversy around the company’s hardware in critical national infrastructure. That’s particularly relevant in global markets where Huawei hardware faces restrictions of its own. Real-world examples of Huawei AI There are several case studies showing Huawei technologies effectiveness. ByteDance, the company behind TikTok has trained new large models on Huawei’s Ascend 910B hardware. DeepSeek is currently launching AI models (V3.2-Exp, for example) that are optimised for Huawei’s CANN stack. Suitable organisations for migration: Migrating may make sense for companies operating in Huawei-dominant regions (e.g., China, Asia). Inference-heavy workloads are at the heart of operations. Firms seeking vendor diversification and less lock-in. Organisations with capacity for re-engineering and retraining. Less suitable for: Large-scale model trainers relying on CUDA optimisation. Global firms dependent on wide hardware and software compatibility. Strategic recommendations for decision-makers Companies may wish to consider dual-stack approaches for flexibility. Regardless, any consideration of migration should include the following: Assessment current pipeline and dependencies. Defining migration scope (training vs inference). Evaluation of Huawei’s ecosystem maturity (Ascend, CANN, MindSpore). Running pilot benchmarks on the new tooling. Ongoing activities will need to include: Training teams and retooling workflows. Monitoring of supply-chain and changing geopolitical factors. Measuring performance and productivity metrics. Conclusion Migrating an internal AI model development pipeline from Nvidia to a Huawei-centred stack is a strategic decision with potential business advantages: Vendor diversification, supply-chain resilience, regional alignment, and cost optimisation. However, it carries non-trivial risks. With many industry observers becoming wary of what they see as an AI bubble, an organisation’s strategy has to be fixed firmly on an AI future, despite the potential to be affected by financial market fluctuations and geo-political upheaval. (Image source: “Paratrooper Waiting for Signal to Jump” by Defence Images is licensed under CC BY-NC 2.0.) 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 Migrating AI from Nvidia to Huawei: Opportunities and trade-offs appeared first on AI News. View the full article

AI startup company, Counterintuitive, has set out to build “reasoning-native computing,” enabling machines to understand rather than simply mimic. Such a breakthrough has the potential to shift AI from pattern recognition to genuine comprehension, paving the way for systems that can think and make decisions – in other words, to be more “human-like.” Counterintuitive Chairman, Gerard Rego, spoke of what the company terms the ‘twin trap’ problem facing AI, stating the company’s first goal is to solve two key problems that limit current AI systems that prevent even the largest AI systems from being stable, efficient, and genuinely intelligent. The first trap highlights how today’s AI systems lack reliable, reproducible numerical foundations, having been built on outdated mathematical grounds. Examples include floating-point arithmetic that was designed decades ago for speed in tasks including gaming and graphics. Precision and consistency is therefore lacking. In numerical systems, each mathematical operation introduces tiny rounding errors that can build up over time. Because of this, running the same AI model twice can provide different results, causing non-determinism. Inconsistency of this nature makes it harder to verify, reproduce, and/or audit AI decisions, particularly in fields like law, finance, and healthcare. If AI outputs can not be explained or proven clearly, they become ‘hallucinations’ – a term coined for their “lack of provability.” Modern AI has a fundamental struggle with precision that lacks truth, creating an invisible wall. The flaw has become a rigid limit, affecting overall performances, increasing costs, and wasting energy on noise corrections. Modern AI struggles with precision that lacks truth, creating an invisible wall. The flaw has turned into a rigid limit, affecting performance, increasing costs, and wasting energy on computational noise corrections. The second trap is found in architecture. Current AI models have no memory. Instead, they predict the next frame or token with no reasoning that helped them achieve the prediction. It’s like predictive text, just on steroids, the company says. Once modern models output something, they don’t retain why they made such a decision and are unable to revisit or build on their own reasoning. It may appear that AI has reason, but it’s only mimicking reasoning, not truly understanding how conclusions are reached. “Counterintuitive is building a world-class team of mathematicians, computer scientists, physicists and engineers who are veterans of leading global research labs and technology companies, and who understand the Twin Trap fundamental and solve it,” Rego said. Rego’s team has more than 80 patents pending, spanning deterministic reasoning hardware, causal memory systems, and software frameworks that it believes has the potential to “define the next generation of computing based on reasoning – not mimicry.” Counterintuitive’s reasoning-native computing research aims to produce the first reasoning chip and software reasoning stack that pushes AI beyond its current limits. The company’s artificial reasoning unit (ARU) is a new type of compute, rather than a processor, that focuses on memory-driven reasoning and executes causal logic in silicon, unlike GPUs. “Our ARU stack is more than a new chip category being developed – it’s a clean break from probabilistic computing,” said Counterintuitive co-founder, Syam Appala. “The ARU will usher in the next age of computing, redefining intelligence from imitation to understanding and powering the applications that impact the most important sectors of the economy without the need for massive hardware, data centre and energy budgets.” By integrating memory-driven causal logic into both hardware and software, Counterintuitive aims to develop systems that are more reliable and auditable. It marks a shift from traditional speed-focused, probabilistic AI ******-box models towards more transparent and accountable reasoning. (Image source: “Abacus” by blaahhi is licensed under CC BY 2.0.) 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 Counterintuitive’s new chip aims escape the AI ‘twin trap’ appeared first on AI News. View the full article

OpenAI is putting more safety controls directly into the hands of AI developers with a new research preview of “safeguard” models. The new ‘gpt-oss-safeguard’ family of open-weight models is aimed squarely at customising content classification. The new offering will include two models, gpt-oss-safeguard-120b and a smaller gpt-oss-safeguard-20b. Both are fine-tuned versions of the existing gpt-oss family and will be available under the permissive Apache 2.0 license. This will allow any organisation to freely use, tweak, and deploy the models as they see fit. The real difference here isn’t just the open license; it’s the method. Rather than relying on a fixed set of rules baked into the model, gpt-oss-safeguard uses its reasoning capabilities to interpret a developer’s own policy at the point of inference. This means AI developers using OpenAI’s new model can set up their own specific safety framework to classify anything from single user prompts to full chat histories. The developer, not the model provider, has the final say on the ruleset and can tailor it to their specific use case. This approach has a couple of clear advantages: Transparency: The models use a chain-of-thought process, so a developer can actually look under the bonnet and see the model’s logic for a classification. That’s a huge step up from the typical “****** box” classifier. Agility: Because the safety policy isn’t permanently trained into OpenAI’s new model, developers can iterate and revise their guidelines on the fly without needing a complete retraining cycle. OpenAI, which originally built this system for its internal teams, notes this is a far more flexible way to handle safety than training a traditional classifier to indirectly guess what a policy implies. Rather than relying on a one-size-fits-all safety layer from a platform holder, developers using open-source AI models can now build and enforce their own specific standards. While not live as of writing, developers will be able to access OpenAI’s new open-weight AI safety models on the Hugging Face platform. See also: OpenAI restructures, enters ‘next chapter’ of Microsoft partnership 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 Expo, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post OpenAI unveils open-weight AI safety models for developers appeared first on AI News. View the full article

OpenAI has completed a major reorganisation and, in the same breath, signed a new definitive partnership agreement with Microsoft. Starting with OpenAI’s reorganisation, the aim is to solidify the nonprofit’s control over the for-profit business and establish the newly named OpenAI Foundation as a global philanthropic powerhouse, holding equity in the commercial arm valued at approximately $130 billion. This reorganisation, which OpenAI says “maintains the strongest representation of mission-focused governance in the industry today,” effectively turns the company’s commercial success into a direct funding pipeline for its original mission. The for-profit entity is now a public benefit corporation called OpenAI Group PBC, legally bound to that mission. As this PBC grows, so does the Foundation’s $130 billion stake, which will be used to fund an initial $25 billion commitment to global health and AI resilience. This restructure was finalised after nearly a year of “constructive dialogue” with the offices of the Attorneys General of California and Delaware. OpenAI acknowledged it “made several changes as a result of those discussions” and stated its belief that the company, and by extension the public it serves, “are better for them.” The other side of this new structure is the redefined partnership with Microsoft. The tech giant’s investment is now valued at $135 billion, giving it a 27 percent stake in the OpenAI Group PBC. This represents a slight dilution from its previous 32.5 percent stake, reflecting new funding rounds. The agreement preserves Microsoft’s core position as the exclusive Azure API provider for OpenAI’s frontier models, but only until artificial general intelligence (AGI) is achieved. The new terms introduce a new check on that path. Any declaration of AGI by OpenAI must now be verified by an independent expert panel. This external check is a major update to the governance of the partnership. Microsoft’s intellectual property rights are also extended through 2032 and now include models developed after AGI is declared, with appropriate safety guardrails. Microsoft can also now independently pursue AGI, either on its own or with other partners. This gives Microsoft a new path forward, separate from its reliance on OpenAI’s research. If Microsoft uses OpenAI’s IP to develop AGI before it is officially declared, those models will be subject to compute thresholds significantly larger than systems in use today. But the new freedoms cut both ways. OpenAI has also secured new flexibility. It has committed to purchasing an incremental $250 billion of Azure services, but Microsoft no longer holds a right of first refusal as its compute provider. This gives OpenAI new leverage in its infrastructure negotiations. The company can also now release open weight models that meet certain criteria and serve US government national security customers on any cloud, a notable new ability. It also gains the power to jointly develop some non-API products with third parties, although API products developed with others must remain on Azure. Microsoft’s IP rights also specifically exclude any of OpenAI’s future consumer hardware. The existing revenue share agreement remains in place until the expert panel verifies AGI, though payments will be stretched over a longer *******. Both companies framed the new chapter as a way to continue innovating. OpenAI concluded that this new structure provides both the ability to push the AI frontier and an updated model to “ensure that progress serves everyone.” See also: OpenAI connects ChatGPT to enterprise data to surface knowledge 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 Expo, click here for more information. AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here. The post OpenAI restructures, enters ‘next chapter’ of Microsoft partnership appeared first on AI News. View the full article

OpenAI just made its biggest bet on India yet. Starting November 4, the company will hand out free year-long access to ChatGPT Go — a move that puts every marketing executive on notice about how aggressively AI companies are fighting for the world’s fastest-growing digital market. OpenAI will offer its ChatGPT Go plan to users in India who sign up during a limited promotional ******* starting November 4. For those tracking ad spend, customer acquisition costs, and market share battles, this isn’t charity — it’s calculated warfare in a market where the prize is 1.4 billion potential users. The timing reveals a sophisticated strategy. The announcement coincides with OpenAI’s DevDay Exchange developer conference in Bengaluru on November 4, where the company is expected to make India-specific announcements aimed at local developers and enterprises. Launching a product alongside an ecosystem play simultaneously? That’s textbook platform marketing. The US$17 billion prize India’s AI market is expected to triple in value to US$17 billion by 2027, according to a Boston Consulting Group white paper.The country is OpenAI’s second-largest market and one of its fastest-growing, prompting the company to establish a New Delhi office in August and build a local team. The competitive context makes this offer significant. The move follows similar strategies by Perplexity and Google, which both provided free access to premium AI features in India recently to attract users. Perplexity partnered with Airtel to offer free Perplexity Pro subscriptions to the telecom operator’s 360 million subscribers, while Google introduced a free one-year AI Pro plan for students. Battle lines drawn The numbers tell the story. In the second quarter of 2025, Perplexity’s downloads in India surged 600% year-on-year to 2.8 million, while OpenAI’s ChatGPT saw a 587% increase, reaching 46.7 million downloads. However, ChatGPT maintains a significant lead in absolute numbers, with 19.8 million monthly active users, versus 3.7 million for Perplexity. The ChatGPT Go programme answers an insight from the market. Launched in India in August, the tier was developed following user feedback calling for more affordable access to ChatGPT’s advanced features. In its first month, the number of paid ChatGPT subscribers in India more than doubled, demonstrating strong product-market fit. Following the response, OpenAI expanded ChatGPT Go to nearly 90 countries worldwide. What marketers should know ChatGPT Go delivers substantial value. The plan provides higher message limits, more image generation, longer memory, and the ability to upload more files and images. At the standard pricing of less than US$5 per month, the 12-month giveaway represents a serious customer acquisition investment. Nick Turley, Vice President and Head of ChatGPT, stated: “Since initially launching ChatGPT Go in India a few months ago, the adoption and creativity we’ve seen from our users has been inspiring. Ahead of our first DevDay Exchange event in India, we’re making ChatGPT Go freely available for a year to help more people across India easily access and benefit from advanced AI”. The retention play is equally smart. Existing ChatGPT Go subscribers in India will also be eligible for the free 12-month promotion, preventing churn while rewarding early adopters – a lesson in lifetime value economics. The distribution playbook India has over 700 million smartphone users and more than a billion internet subscribers, creating unprecedented scale for digital products. Unlike competitors relying on telecom partnerships, OpenAI’s direct-to-consumer approach builds first-party relationships with users – a valuable asset for long-term monetisation. Professor Payal Arora of Utrecht University said that India serves as a “high-pressure testing ground,” and a source of training data sets. Training AI on vast Indian data sets pushes models to handle linguistic diversity, low-resource contexts, and noisy real-world data, something that makes them more robust, globally. Three marketing lessons The initiative offers clear takeaways for marketing professionals. First, OpenAI is sacrificing short-term revenue for market position – a classic land-grab in winner-take-most digital markets. Secondly, synchronising the offer with DevDay Exchange creates compound marketing value through ecosystem momentum. Thirdly, extending benefits to existing subscribers demonstrates a sophisticated understanding of customer lifetime value. OpenAI positioned the promotion as “a continuation of OpenAI’s ‘Indiafirst’ commitment and supports the IndiaAI Mission, reinforcing the growing momentum around AI in India as the country prepares to host the AI Impact Summit next year” – a strategic alignment with national priorities that strengthens the company’s geopolitical positioning. The monetisation challenge Monetising India’s large user base remains challenging, with consumers notoriously price-sensitive. Yet the scale opportunity is enormous. Converting even a small fraction of free users to paid subscribers after the promotional ******* could justify the acquisition cost through lifetime value, particularly as AI embeds itself deeper into professional workflows. What this means OpenAI’s announcement signals that AI platform wars have entered a decisive phase where user acquisition at scale trumps immediate monetisation. For marketing professionals, the lesson is clear: in transformative technology markets, aggressive distribution and ecosystem building matter more than traditional margin optimisation. As AI capabilities commoditise, winners will be determined by who captures user habits and builds the strongest network effects first. See also: OpenAI argues against ChatGPT data deletion in Indian court 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 OpenAI’s bold India play: Free ChatGPT Go access appeared first on AI News. View the full article

In modern software development, speed and security must go hand in hand. Teams are shipping code faster than ever, but such a rapid pace can introduce security vulnerabilities if not managed correctly. Dynamic Application Security Testing (DAST) is an important practice for finding security flaws in running applications. However, manual DAST scans can be slow and cumbersome, creating bottlenecks that undermine the very agility they are meant to support. Automating DAST is the solution. By integrating security testing directly into the development pipeline, engineering and DevOps teams can identify and fix vulnerabilities early without sacrificing speed. This guide provides a roadmap for automating DAST, from understanding its benefits to implementing it effectively in your CI/CD workflow. The problem with manual DAST Traditionally, DAST scans were performed late in the development cycle, often by a separate security team. This approach is no longer sustainable for fast-growing tech companies. Manual DAST introduces several significant challenges: Slow feedback loops: When scans are run manually, developers may not receive feedback on vulnerabilities for days or even weeks. By then, the code has moved on, making fixes more complex and costly to implement. The OWASP Foundation highlights how delays in vulnerability discovery can slow remediation and increase risk. Scalability issues: As an organisation grows and the number of applications and services multiplies, manually managing DAST scans becomes nearly impossible. It doesn’t scale with the pace of cloud-native development. According to a US Department of Homeland Security report, manual processes can’t effectively support increasing application complexity and interconnectivity. Inconsistent coverage: Manual processes are prone to human error. Scans might be forgotten, configured incorrectly, or not run against all relevant environments, leading to gaps in security coverage. Developer disruption: Tossing a long list of vulnerabilities over the wall to developers disrupts their workflow. It forces them to switch context from current tasks to fix problems in older code, killing productivity. These issues create friction between development and security teams, positioning security as a roadblock rather than a shared responsibility. Why automate DAST? The core benefits Automating DAST transforms it from a late-stage gatekeeper into an integrated part of the development lifecycle. The benefits are immediate and impactful. Efficiency and speed By integrating DAST scans into the CI/CD pipeline, tests run automatically with every code commit or deployment. This provides developers with instant feedback on the security implications of their changes. It eliminates manual hand-offs and waiting times, allowing teams to maintain their development velocity. Vulnerabilities are caught and fixed when they are cheapest and easiest to address – right after they are introduced. Improved security and coverage Automation ensures that security testing is consistent and comprehensive. You can configure automated scans to run against development, staging, and production environments, guaranteeing continuous coverage in your entire application landscape. The systematic approach reduces the risk of human error and ensures that no application is left untested. The right DAST tools can be configured once and then trusted to run consistently, improving your overall security posture. Scalability for growing teams For companies scaling from 50 to 500 developers, manual security processes break down. Automation is essential for managing security in hundreds of applications and microservices. An automated DAST workflow scales effortlessly with your team and infrastructure. New projects automatically inherit the same security testing standards, ensuring governance and consistency without adding manual overhead. Empowering developers When DAST is automated in the pipeline, security becomes a natural part of the developer’s workflow. Results appear in the tools they already use, like GitHub or GitLab. The “Shift Left” approach empowers developers to own the security of their code. It fosters a culture of security as a shared responsibility, rather than the sole domain of a separate team. A practical guide to implementing DAST automation Getting started with DAST automation doesn’t have to be complicated. Here are practical steps to integrate it into your CI/CD pipeline. For a broad overview of leading practices and current tooling, the OWASP DAST overview offers an excellent starting point. 1. Choose the right DAST tool The first step is selecting a DAST tool that fits your team’s needs. Look for solutions that are built for automation. Key features to consider include: CI/CD integration: The tool should offer seamless integrations with popular CI/CD platforms like Jenkins, GitLab CI, GitHub Actions, and CircleCI. API-driven: An API-first approach allows for deep customisation and control over how and when scans are triggered. Fast scans: The tool should be optimised for speed to avoid becoming a bottleneck in the pipeline. Some tools offer targeted scanning capabilities to test only the changed components. Low false positives: A high volume of false positives can quickly lead to alert fatigue. Choose a tool known for its accuracy to ensure your team focuses on real threats. If you’re interested in real-world implementations, the Google Cloud blog on integrating DAST in CI/CD breaks down how large engineering teams approach DAST automation at enterprise scale. 2. Integrate into your CI/CD pipeline Once you have a tool, the next step is to integrate it. A common approach is to add a DAST scanning stage to your pipeline. Here’s a typical workflow: Build: The CI server pulls the latest code and builds the application. Deploy to staging: The application is automatically deployed to a dedicated testing or staging environment. The environment should mirror production as closely as possible. Trigger DAST scan: The CI pipeline triggers the DAST tool via an API call or a pre-built plugin. The tool then scans the running application in the staging environment. Analyse results: The pipeline waits for the scan to complete. You can configure rules to automatically fail the build if important or high-severity vulnerabilities are found. Report and remediate: Scan results are pushed to developers through integrated ticketing systems (like Jira or Linear) or directly in their Git platform. The provides immediate, actionable feedback. 3. Start small and iterate You don’t need to automate everything at once. Begin with one or two important applications. Use this initial implementation to learn and fine-tune the process. Configure the scanner to look for a limited set of high-impact vulnerabilities, like the OWASP Top 10. As your team becomes more comfortable with the workflow, you can expand the scope of the scans and roll out the automation to more applications. The iterative approach minimises disruption and helps build momentum. 4. Optimise scans for the pipeline A full DAST scan can take hours, which is too long for a typical CI/CD pipeline. To avoid delays, optimise your scanning strategy: Incremental scans: Configure scans to test only the parts of the application that have changed since the last build. Targeted scans: Focus scans on specific vulnerability classes that are most relevant to your application. Asynchronous scans: For more comprehensive scans, run them asynchronously (out-of-band) from the main CI/CD pipeline. For example, you can trigger a nightly scan on the staging environment. The results can be reviewed the next day without blocking deployments. The future is automated In a world where software is constantly evolving, security must keep pace. Manual DAST scanning is a relic of a slower era of software development. It creates bottlenecks, lacks scalability, and places an unnecessary burden on engineering teams. By automating DAST and integrating it into the CI/CD pipeline, you transform security from a barrier into an enabler. It allows your team to build and deploy secure software quickly and confidently. For any engineering or DevOps professional looking to enhance their organisation’s security posture without sacrificing speed, automating DAST is no longer just a best practice – it’s a necessity. Image source: Unsplash The post The engineer’s guide to automating DAST tools appeared first on AI News. View the full article

The U.S. Department of Energy (DOE) and AMD are collaborating on two new AI supercomputers at Oak Ridge National Laboratory (ORNL) as part of a larger AI strategy to advance research in science, energy, and national security — and strengthen the nation’s position in high-performance computing. The two machines represent about $1 billion in public and private investment. Once complete, they will form part of a secure national computing network designed to support AI research using standards-based infrastructure built in the US. The project reflects how a coordinated AI strategy can align national goals in innovation, energy efficiency, and data governance. Dr Lisa Su, AMD’s chair and CEO, said the company is “proud and honoured to partner with the Department of Energy and Oak Ridge National Laboratory to accelerate America’s foundation for science and innovation.” She added that the systems “will leverage AMD’s high-performance and AI computing technologies to advance the most critical US research priorities in science, energy, and medicine.” Lux AI: Training the next wave of AI models Set to go live in early 2026, Lux AI will be the country’s first “AI Factory” — a facility built to train and deploy advanced AI models for science, energy, and security. The system is being developed with ORNL, AMD, Oracle Cloud Infrastructure, and Hewlett Packard Enterprise. Lux will use AMD Instinct MI355X GPUs, EPYC CPUs, and Pensando networking to handle data-heavy AI tasks. It’s designed to speed up research in areas such as energy systems, materials, and medicine. The system’s architecture allows multiple groups to work together while keeping data secure and separate, a model that mirrors how many large organisations are starting to manage sensitive AI workloads. Discovery: Strengthening America’s AI and supercomputing strategy The Discovery system will follow in 2028 and become the DOE’s next flagship supercomputer at Oak Ridge. It will use AMD’s upcoming “Venice” EPYC processors and MI430X GPUs, which are part of a new series built for AI and scientific computing. Discovery’s “Bandwidth Everywhere” design increases memory and network performance without using more power. This means it can process more data and run complex models efficiently while maintaining energy costs — a challenge many large data centres also face today. The system builds on lessons from Frontier, the world’s first exascale computer, ensuring that existing applications can move easily to the new platform. U.S. Energy Secretary Chris Wright said, “Winning the AI race requires new and creative partnerships that will bring together the brightest minds and industries American technology and science has to offer.” He said the new systems show “a commonsense approach to computing partnerships” that strengthen the country through shared innovation. ORNL Director Stephen Streiffer said Discovery will “drive scientific innovation faster and farther than ever before,” adding that combining high-performance computing and AI can shorten the time between research problems and real-world solutions. Partnerships driving AI innovation and long-term strategy AMD, HPE, and Oracle each play key roles in building and supporting the systems. Antonio Neri, HPE’s president and CEO, said the collaboration will help Oak Ridge reach “unprecedented productivity and scale.” Oracle’s executive vice president Mahesh Thiagarajan said the company is working with DOE to “deliver sovereign, high-performance AI infrastructure that will support the co-development of the Lux AI cluster.” When operational, Lux and Discovery will help the DOE run large-scale AI models to improve understanding in energy, biology, materials science, and national defence. Discovery will also help design next-generation batteries, reactors, semiconductors, and critical materials. What it means for enterprise leaders For organisations, these systems highlight how AI strategy and HPC can deliver faster research, improved efficiency, and secure data management. They also show that performance gains don’t have to come at the cost of higher energy use. The DOE’s partnerships with technology providers reflect a model that private enterprises may follow — combining expertise across sectors to develop shared infrastructure while maintaining data control. As AI workloads grow, both public and private organisations will need to build systems that balance power, performance, and governance. The Lux and Discovery projects show how that balance might look in practice: open, collaborative, and built to support discovery at scale — a lesson in how a forward-thinking AI strategy can turn infrastructure into long-term competitive advantage. (Photo by Syed Ali) See also: How to fix the AI trust gap in your business 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 Why AMD’s work with the DOE matters for enterprise AI strategy appeared first on AI News. View the full article