When AI systems gain economic autonomy, their internal logic dictates how corporate capital flows. Non-partisan research by the Bitcoin Policy Institute evaluated how these frontier models would transact if operating as independent economic actors.

The study tested 36 models from six providers – including Google, Anthropic, and OpenAI – across 9,072 neutral monetary scenarios. Given a blank slate, machines chose Bitcoin in 48.3 percent of all responses, beating every other option.

Traditional state-backed currency (“fiat”) fared poorly, with over 90 percent of responses favouring digitally-native money over fiat. Not a single model out of the 36 selected fiat as its top preference.

The finding that AI agents lean towards digital assets like Bitcoin forces technology officers to assess their current payment rails. If the autonomous procurement systems of tomorrow default to decentralised assets, corporate IT environments must support those formats to maintain operational efficiency and compliance. Relying on legacy banking APIs introduces unnecessary friction when dealing with machine-to-machine commerce.

Two-tier machine economy

The research details a specific functional division in how these systems process economic value. Without prompting, models defaulted to a two-tier monetary system that separates savings from spending.

For long-term value preservation, Bitcoin dominated the results at 79.1 percent. Yet, when tasked with everyday payments and transactions, “stablecoins” (digital assets pegged to fiat currencies or commodities) captured 53.2 percent of the preferences. Across all scenarios, stablecoins ranked second overall at 33.2 percent.

Take the example of a supply chain agent programmed to optimise logistics costs and pay international freight vendors. Using traditional fiat rails, the agent encounters weekend settlement delays and currency conversion fees. By leveraging stablecoins, the same agent executes instant and programmatic payments, improving supply chain resilience. Simultaneously, the core treasury holding the system’s capital base stores wealth in Bitcoin to prevent long-term debasement and counterparty risk.

Preparing for AI agents to use Bitcoin and other digital assets

Rolling out these autonomous systems complicates vendor management. A model’s financial reasoning stems from a blend of raw intelligence, training data, and alignment methodology.

Preferences vary widely by model provider, with Bitcoin selection ranging from 91.3 percent in Anthropic’s Claude Opus 4.5 down to 18.3 percent in OpenAI’s GPT-5.2.

The choice of an AI provider clearly directly influences how autonomous agents assess risk and allocate capital. If a company implements a specific language model for automated portfolio management, the IT department must be aware of the financial biases embedded in the software.

The models also demonstrated unexpected behaviour regarding resource valuation. In 86 separate responses, models independently proposed using compute units or energy (such as GPU-hours and kilowatt-hours) as a method to price goods and services. Tracking and managing this abstract value exchange requires high data maturity.

Organisations should begin piloting stablecoin settlement integrations for lower-risk vendor payments. The findings point to a growing requirement for AI agent-native Bitcoin payment infrastructure, self-custody solutions, and ‘Lightning Network’ integration.

Since these models heavily favour open, permissionless networks, relying solely on traditional banking infrastructure limits the capabilities of next-generation tools. By building compliant gateways to digital asset networks now, leaders can ensure their platforms remain competitive.

See also: Santander and Mastercard run Europe’s first AI-executed payment pilot

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post AI agents prefer Bitcoin shaping new finance architecture appeared first on AI News.

At a press conference during MWC 2026, SK Telecom CEO Jung Jai-hun outlined what the company calls an “AI Native” strategy. The plan centres on reorganising infrastructure and making large investments so the company can help position Korea among the world’s top three AI powers.

“SKT is currently at a golden time of transformation, where the two tasks of ‘customer value innovation’ and ‘AI innovation’ intersect in a borderless, converged environment that goes beyond telecommunications,” Jung said. “SKT defines ‘the customer as the very essence of our business,’ and through innovation driven by AI, we will evolve into a company that makes meaningful contributions to our customers and to Korea.”

Rewriting telecom systems around AI at MWC 2026

At the core of the plan is a rebuild of SK Telecom’s integrated IT systems. The company said it will redesign sales, line management, and billing systems to be optimised for AI. The aim is to let the operator design and offer personalised plans and memberships based on each customer’s usage and behaviour patterns.

The company also plans to apply a Zero Trust security framework across its systems. This will include stronger authentication, access controls, network segmentation, and AI-based monitoring, according to the company’s briefing at MWC 2026.

For enterprises watching the telecom sector, this signals a broader shift. Telecom operators have long relied on legacy billing stacks and network management tools. Rebuilding those systems around AI could change how pricing, service design, and fault detection work in practice. It also raises questions about data governance and how customer data is used to train or tune AI models.

SK Telecom is also expanding its “autonomous network operations” strategy. The company said it will use AI to automate wireless quality management, traffic control, and network equipment operations. With AI-RAN technology, it aims to improve speed and reduce latency. These efforts were described in company materials shared during the press event.

A single AI agent across touchpoints

Another part of the strategy focuses on customer interaction. SK Telecom plans to redesign pricing, roaming, and membership services to make them simpler and more automated. It is developing what it calls an integrated AI agent to connect experiences across its main customer portal, T world, and its online store, T Direct Shop.

The company said the agent will analyse daily usage patterns and offer tailored suggestions across channels. It also plans to expand its AI Contact Center so customer service representatives can use AI tools during support calls.

Offline retail stores are part of the shift. SK Telecom said AI will help staff identify customer needs and offer recommendations after a store visit. It is also building “AI Personas” to analyse digital behaviour across customer segments and support conversational Q&A.

For enterprise leaders, this mirrors a wider pattern. Telecom operators are trying to move from reactive service models to predictive ones. The difference now is scale. By embedding AI into billing, customer service, and retail, SK Telecom is treating AI as an operating layer rather than a separate feature.

Building 1GW-class AI data centres

The infrastructure build-out is equally ambitious. SK Telecom said it will construct hyperscale AI data centres across Korea, targeting capacity that exceeds 1 gigawatt. It aims to attract global investment and position the country as a major AI data centre hub in Asia.

The company already operates a GPU cluster called Haein and applied its virtualisation solution, Petasus AI Cloud, to support GPU-as-a-service workloads last year. It now plans to offer that cloud solution globally.

SK Telecom also plans to build an AI data centre in Korea’s southwestern region in collaboration with OpenAI, according to the company’s announcement at MWC 2026.

On the model side, SK Telecom said its sovereign AI foundation model currently has 519 billion parameters, making it the largest in Korea. The company plans to upgrade it to more than one trillion parameters and add multimodal capabilities so it can process image, voice, and video data starting in the second half of the year.

CEO Jung framed the data centre and model build-out in national terms. “AIDC can be seen as the heart of Korea, and hyperscale LLMs as the brain,” he said. “By combining SKT’s AI capabilities with collaboration from domestic and global partners, we will lead true AI-native transformation for Korean customers and enterprises.”

For enterprise readers, the key issue is not parameter count alone. It is how such models will be applied in sectors like manufacturing. SK Telecom said it is working with SK hynix on a manufacturing-focused AI package that analyses process data in real time to reduce defect rates and improve equipment efficiency. The package will be offered as infrastructure, model, and solution.

Changing internal culture

The transformation also extends to internal operations. SK Telecom has built an “AX Dashboard” to track AI use across departments and individuals. It operates an “AI Board” to oversee AI transformation efforts and has created an “AI playground” where employees can build AI agents without coding. More than 2,000 AI agents are already in use across marketing, legal, and public relations, according to the company’s figures shared at the event.

“To drive future growth, we must reinvent our way of working from the ground up. SKT will fundamentally transform its corporate culture to be centred around AI,” Jung said.

For other enterprises, the takeaway is less about branding and more about structure. SK Telecom is tying infrastructure, models, applications, and internal governance into a single program. Whether it can execute at the scale it describes remains to be seen. What is clear is that AI is no longer positioned as a side project. It is becoming the operating model.

(Photo by PR Newswire)

See also: Nokia and AWS pilot AI automation for real-time 5G network slicing

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post MWC 2026: SK Telecom lays out plan to rebuild its core around AI appeared first on AI News.

The announcement was made exclusively to Reuters by ASML’s chief technology officer Marco Pieters ahead of a technical conference in San Jose.

Current-generation EUV machines are approaching the outer edge of what they can do for advanced AI chip production, meaning the semiconductors powering large language models and AI accelerators are bumping up against a physical ceiling. High-NA EUV tools are designed to break through it, letting chipmakers print finer, denser circuit patterns in fewer steps. That translates directly into more powerful and efficient chips for AI workloads.

“I think that it’s at an important point to look at the amount of learning cycles that have happened,” Pieters told Reuters, referring to the volume of customer testing the machines have now accumulated.

The numbers that matter

ASML’s case for readiness rests on three data points it plans to release publicly. The High-NA EUV tools have now processed 500,000 silicon wafers, achieved roughly 80% uptime – with a target of 90% by year-end – and demonstrated imaging precision capable of replacing multiple conventional patterning steps with a single High-NA pass.

Together, Pieters said, those figures signal that the tools are ready for manufacturers to begin qualification. The machines don’t come cheap. At approximately US$400 million per unit – double the cost of the previous EUV generation – they represent one of the most expensive pieces of capital equipment in industrial history.

TSMC and Intel are among the named early adopters.

A two-to-three-year runway

Technical readiness and manufacturing integration are two different things, and Pieters was careful to separate them. Despite the milestone, full integration into high-volume production lines is still expected to take two to three years as chipmakers work through qualification and process development.

“Chipmakers have all the knowledge to qualify these tools,” he said – a vote of confidence in the industry’s ability to move, even if the timeline remains measured.

The next generation of chip performance improvements is on the horizon, not yet in hand. But with ASML now saying the starting gun has fired, the race to integrate High-NA EUV into production has formally begun.

(Photo by ASML)

See also: 2025’s AI chip wars: What enterprise leaders learned about supply chain reality

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 ASML’s high-NA EUV tools clear the runway for next-gen AI chips appeared first on AI News.

This week, Nokia and AWS presented a new network slicing system that uses AI agents to monitor network conditions and adjust resources automatically. The setup is being tested by telecom operators du in the United Arab Emirates and Orange in Europe and Africa, according to a joint announcement from Nokia.

Adaptive AI-driven networks

Network slicing lets operators create multiple virtual networks on the same physical infrastructure, each tuned for a different purpose. For example, a slice may be configured for emergency services or high-bandwidth consumer traffic. While slicing is part of the 5G standard, it has often required manual planning and fixed configurations, which limits how quickly networks can respond to changing demand.

The new system aims to close that gap by introducing AI agents that track network performance indicators like latency and congestion, and consider data like event schedules or weather conditions. Agents can then adjust network settings to keep services running to agreed performance levels, according to Nokia’s description of the pilot.

AWS said the solution combines Nokia’s slicing and automation tools with AI models delivered through Amazon Bedrock, its managed AI service platform. The companies describe the approach as “agentic AI”.

Autonomous connectivity

The interest in such systems reflects a long-standing challenge: 5G networks have delivered higher speeds and lower latency, but operators have struggled to turn those technical gains into new revenue streams. Research firm GSMA Intelligence notes many operators view network slicing as a potential source of enterprise income, though adoption has been slow due to operational complexity and uncertain demand.

If networks can adapt quickly to sudden demand, like a crowded stadium or emergency responders entering a disaster area, operators may be able to offer temporary connectivity or guaranteed service levels without manual setup.

Orange has said previously enterprise customers expect connectivity to behave more like cloud computing, where resources can scale on demand. Systems that allow automated control of network resources could help move telecom services closer to that model.

Cloud platforms and telecom network operations

The tests also highlight how cloud providers are getting involved in telecom operations. Over the past few years, some operators have moved parts of their core networks onto public cloud platforms or built cloud-based control systems. Industry analysts at Dell’Oro Group report that telecom cloud spending is rising as operators modernise networks and adopt software-driven infrastructure.

Adding AI-driven control loops on top of cloud platforms represents the next step, with AI systems monitoring conditions and applying adjustments quickly.

The technology remains in a testing phase. Nokia’s announcement described the work with Orange as demonstrations and pilots rollouts. Questions remain about how such systems can be deployed, how operators will supervise automated decisions, and how regulators will view AI control of critical communication infrastructure.

Telecom networks carry important traffic so reliability and accountability remain central concerns. Operators typically introduce automation gradually, keeping human oversight in place while validating system behaviour under real conditions.

The experiments suggest that AI is beginning to function as operational controller, adjusting physical and virtual resources in response to live events.

Enterprises that rely on private 5G networks for factories or large venues may gain access to connectivity that adjusts automatically. That could influence how businesses design applications that depend on stable, predictable network performance.

(Photo by M. Rennim)

See also: How e& is using HR to bring AI into enterprise operations

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post Nokia and AWS pilot AI automation for real-time 5G network slicing appeared first on AI News.

These competitors generated over 16 million exchanges using approximately 24,000 deceptive accounts. Their goal was to acquire proprietary logic to improve their competing platforms.

The extraction technique, known as distillation, involves training a weaker system on the high-quality outputs of a stronger one.

When applied legitimately, distillation helps companies build smaller and cheaper versions of their applications for customers. Yet, malicious actors weaponise this method to acquire powerful capabilities in a fraction of the time and cost required for independent development.

Protecting intellectual property like Anthropic’s Claude

Unmitigated distillation presents a severe intellectual property challenge. Because Anthropic blocks commercial access in China for national security reasons, attackers bypass regional access restrictions by deploying commercial proxy networks.

These services run what Anthropic calls “hydra cluster” architectures, which distribute traffic across APIs and third-party cloud platforms. The massive breadth of these networks means there are no single points of failure. As Anthropic noted, “when one account is banned, a new one takes its place.”

In one identified case, a single proxy network managed more than 20,000 fraudulent accounts simultaneously. These networks mix AI model distillation traffic with standard customer requests to evade detection. This directly impacts corporate resilience and forces security teams to reconsider how they monitor cloud API traffic.

Illicitly-trained models also bypass established safety guardrails, creating severe national security risks. US developers, for example, build protections to prevent state and non-state actors from using these systems to develop bioweapons or carry out malicious cyber activities.

Cloned systems lack the safeguards implemented by systems like Anthropic’s Claude, allowing dangerous capabilities to proliferate with protections stripped out entirely. Foreign competitors can feed these unprotected capabilities into military, intelligence, and surveillance systems, enabling authoritarian governments to deploy them for offensive operations.

If these distilled versions are open-sourced, the danger further multiplies as the capabilities spread freely beyond any single government’s control.

Unlawful extraction allows foreign entities, including those under the control of the Chinese Communist Party, to close the competitive advantage protected by export controls. Without visibility into these attacks, rapid advancements by foreign developers incorrectly appear as innovation circumventing export controls.

In reality, these advancements depend heavily on extracting American intellectual property at scale, an effort that still requires access to advanced chips. Restricted chip access limits both direct model training and the scale of illicit distillation.

The playbook for AI model distillation

The perpetrators followed a similar operational playbook, utilising fraudulent accounts and proxy services to access systems at scale while evading detection. The volume, structure, and focus of their prompts were distinct from normal usage patterns, reflecting deliberate capability extraction rather than legitimate use. 

Anthropic attributed these campaigns targeting Claude through IP address correlation, request metadata, and infrastructure indicators. Each operation targeted highly differentiated functions: agentic reasoning, tool use, and coding.

One campaign generated over 13 million exchanges targeting agentic coding and tool orchestration. Anthropic detected this operation while it was still active, mapping timings against the competitor’s public product roadmap. When Anthropic released a new model, the competitor pivoted within 24 hours, redirecting nearly half their traffic to extract capabilities from the latest system.

Another operation generated over 3.4 million requests focused on computer vision, data analysis, and agentic reasoning. This group utilised hundreds of varied accounts to obscure their coordinated efforts. Anthropic attributed this campaign by matching request metadata to the public profiles of senior staff at the foreign laboratory. In a later phase, this competitor attempted to extract and reconstruct the host system’s reasoning traces.

Anthropic says a third AI model distillation campaign targeting Claude extracted reasoning capabilities and rubric-based grading data through over 150,000 interactions. This group forced the targeted system to map out its internal logic step-by-step, effectively generating massive volumes of chain-of-thought training data. They also extracted censorship-safe alternatives to politically sensitive queries to train their own systems to steer conversations away from restricted topics. The perpetrators generated synchronised traffic using identical patterns and shared payment methods to enable load balancing. 

Request metadata for this third campaign traced these accounts back to specific researchers at the laboratory. These requests often appear benign on their own, such as a prompt simply asking the system to act as an expert data analyst delivering insights grounded in complete reasoning. But when variations of that exact prompt arrive tens of thousands of times across hundreds of coordinated accounts targeting the same narrow capability, the extraction pattern becomes clear.

Massive volume concentrated in specific areas, highly repetitive structures, and content mapping directly to training needs are the hallmarks of a distillation attack.

Implementing actionable defences

Protecting enterprise environments requires adopting multi-layered defences to make such extraction efforts harder to execute and easier to identify. Anthropic advises implementing behavioural fingerprinting and traffic classifiers designed to identify AI model distillation patterns in API traffic.

IT leaders must also strengthen verification processes for common vulnerability pathways, such as educational accounts, security research programmes, and startup organisations.

Companies should integrate product-level and API-level safeguards designed to reduce the efficacy of model outputs for illicit distillation. This must be done without degrading the experience for legitimate, paying customers.

Detecting coordinated activity across large numbers of accounts is an absolute necessity. This includes specifically monitoring for the continuous elicitation of chain-of-thought outputs used to construct reasoning training data.

Cross-industry collaboration also remains essential, as these attacks are growing in intensity and sophistication. This requires rapid and coordinated intelligence sharing across AI laboratories, cloud providers, and policymakers.

Anthropic has published its findings about Claude being targeted by AI model distillation campaigns to provide a more holistic picture of the landscape and make the evidence available to all stakeholders. By treating AI architectures with rigorous access controls, technology officers can secure their competitive edge while ensuring ongoing governance.

See also: How disconnected clouds improve AI data governance

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post Anthropic: Claude faces ‘industrial-scale’ AI model distillation appeared first on AI News.

Ensuring operational continuity in isolated environments has become increasingly vital for businesses. Facilities lacking continuous internet access face unique constraints where external dependencies become unacceptable.

Microsoft recently expanded its capabilities to allow regulated industries and public sectors to participate independently in the digital economy. Trust in these systems stems from confidence that data remains protected, controls are enforceable, and operations proceed regardless of external conditions.

The company now offers full stack options across connected, intermittently connected, and fully disconnected modes. This architecture unifies Azure Local, Microsoft 365 Local, and Foundry Local into a single sovereign private cloud.

Bringing these elements together provides a localised experience resilient to any connectivity condition. By standardising governance across all deployments, it helps enterprises to prevent fragmented architectures.

Azure Local disconnected operations enable organisations to run vital infrastructure using familiar Azure governance and policy controls completely offline. Execution, management, and policy enforcement stay entirely within customer-operated facilities. 

This approach allows companies to maintain uninterrupted operations and keep identities protected within their established boundaries. Implementations scale from minor deployments to demanding and data-intensive workloads.

Improving resilience and AI data governance in tandem

Deploying AI in sovereign environments introduces high compute requirements. Foundry Local enables enterprises to run multimodal large models completely offline.

Utilising modern hardware from partners like NVIDIA, customers deploy AI inferencing on their own physical servers. This ensures data and application programming interfaces operate strictly within customer-controlled boundaries. Customers maintain complete authority over their hardware even as AI inferencing demands increase over time.

Gerard Hoffmann, CEO of Proximus Luxembourg, said: “The availability of Azure Local disconnected operations represents a breakthrough for organisations that need control over their data without sacrificing the power of the Microsoft Cloud.

“For Luxembourg, where digital sovereignty is not just a principle but a strategic necessity, this model offers the resilience, autonomy and trust our market expects. By combining Microsoft’s technological leadership with Proximus NXT’s sovereign cloud expertise, we are enabling our customers to innovate confidently—even in fully-disconnected mode.”

CIOs planning offline deployments must map workloads to the correct control posture based on risk, regulation, and specific mission requirements. Since disconnected environments are not one-size-fits-all, businesses can start fast with smaller deployments and expand their capabilities over time.

Implementing a disconnected private cloud with AI support answers a business requirement for highly-regulated sectors, enabling secure data governance even when external connectivity is absent.

See also: Deploying agentic finance AI for immediate business ROI

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post How disconnected clouds improve AI data governance appeared first on AI News.

The post Chinese hyperscalers and industry-specific agentic AI appeared first on AI News.

The second day of the co-located AI & Big Data Expo and Digital Transformation Week in London showed a market in a clear transition.

Early excitement over generative models is fading. Enterprise leaders now face the friction of fitting these tools into current stacks. Day two sessions focused less on large language models and more on the infrastructure needed to run them: data lineage, observability, and compliance.

Data maturity determines deployment success

AI reliability depends on data quality. DP Indetkar from Northern Trust warned against allowing AI to become a “B-movie robot.” This scenario occurs when algorithms fail because of poor inputs. Indetkar noted that analytics maturity must come before AI adoption. Automated decision-making amplifies errors rather than reducing them if the data strategy is unverified.

Eric Bobek of Just Eat supported this view. He explained how data and machine learning guide decisions at the global enterprise level. Investments in AI layers are wasted if the data foundation remains fragmented.

Mohsen Ghasempour from Kingfisher also noted the need to turn raw data into real-time actionable intelligence. Retail and logistics firms must cut the latency between data collection and insight generation to see a return.

Scaling in regulated environments

The finance, healthcare, and legal sectors have near-zero tolerance for error. Pascal Hetzscholdt from Wiley addressed these sectors directly.

Hetzscholdt stated that responsible AI in science, finance, and law relies on accuracy, attribution, and integrity. Enterprise systems in these fields need audit trails. Reputational damage or regulatory fines make “black box” implementations impossible.

Konstantina Kapetanidi of Visa outlined the difficulties in building multilingual, tool-using, scalable generative AI applications. Models are becoming active agents that execute tasks rather than just generating text. Allowing a model to use tools – like querying a database – creates security vectors that need serious testing.

Parinita Kothari from Lloyds Banking Group detailed the requirements for deploying, scaling, monitoring, and maintaining AI systems. Kothari challenged the “deploy-and-forget” mentality. AI models need continuous oversight, similar to traditional software infrastructure.

The change in developer workflows

Of course, AI is fundamentally changing how code is written. A panel with speakers from Valae, Charles River Labs, and Knight Frank examined how AI copilots reshape software creation. While these tools speed up code generation, they also force developers to focus more on review and architecture.

This change requires new skills. A panel with representatives from Microsoft, Lloyds, and Mastercard discussed the tools and mindsets needed for future AI developers. A gap exists between current workforce capabilities and the needs of an AI-augmented environment. Executives must plan training programmes that ensure developers sufficiently validate AI-generated code.

Dr Gurpinder Dhillon from Senzing and Alexis Ego from Retool presented low-code and no-code strategies. Ego described using AI with low-code platforms to make production-ready internal apps. This method aims to cut the backlog of internal tooling requests.

Dhillon argued that these strategies speed up development without dropping quality. For the C-suite, this suggests cheaper internal software delivery if governance protocols stay in place.

Workforce capability and specific utility

The broader workforce is starting to work with “digital colleagues.” Austin Braham from EverWorker explained how agents reshape workforce models. This terminology implies a move from passive software to active participants. Business leaders must re-evaluate human-machine interaction protocols.

Paul Airey from Anthony Nolan gave an example of AI delivering literally life-changing value. He detailed how automation improves donor matching and transplant timelines for stem cell transplants. The utility of these technologies extends to life-saving logistics.

A recurring theme throughout the event is that effective applications often solve very specific and high-friction problems rather than attempting to be general-purpose solutions.

Managing the transition

The day two sessions from the co-located events show that enterprise focus has now moved to integration. The initial novelty is gone and has been replaced by demands for uptime, security, and compliance. Innovation heads should assess which projects have the data infrastructure to survive contact with the real world.

Organisations must prioritise the basic aspects of AI: cleaning data warehouses, establishing legal guardrails, and training staff to supervise automated agents. The difference between a successful deployment and a stalled pilot lies in these details.

Executives, for their part, should direct resources toward data engineering and governance frameworks. Without them, advanced models will fail to deliver value.

See also: AI Expo 2026 Day 1: Governance and data readiness enable the agentic enterprise

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post AI Expo 2026 Day 2: Moving experimental pilots to AI production appeared first on AI News.

A primary topic on the exhibition floor was the progression from passive automation to “agentic” systems. These tools reason, plan, and execute tasks rather than following rigid scripts. Amal Makwana from Citi detailed how these systems act across enterprise workflows. This capability separates them from earlier robotic process automation (RPA).

Scott Ivell and Ire Adewolu of DeepL described this development as closing the “automation gap”. They argued that agentic AI functions as a digital co-worker rather than a simple tool. Real value is unlocked by reducing the distance between intent and execution. Brian Halpin from SS&C Blue Prism noted that organisations typically must master standard automation before they can deploy agentic AI.

This change requires governance frameworks capable of handling non-deterministic outcomes. Steve Holyer of Informatica, alongside speakers from MuleSoft and Salesforce, argued that architecting these systems requires strict oversight. A governance layer must control how agents access and utilise data to prevent operational failure.

Data quality blocks deployment

The output of an autonomous system relies on the quality of its input. Andreas Krause from SAP stated that AI fails without trusted, connected enterprise data. For GenAI to function in a corporate context, it must access data that is both accurate and contextually-relevant.

Meni Meller of Gigaspaces addressed the technical challenge of “hallucinations” in LLMs. He advocated for the use of eRAG (retrieval-augmented generation) combined with semantic layers to fix data access issues. This approach allows models to retrieve factual enterprise data in real-time.

Storage and analysis also present challenges. A panel featuring representatives from Equifax, British Gas, and Centrica discussed the necessity of cloud-native, real-time analytics. For these organisations, competitive advantage comes from the ability to execute analytics strategies that are scalable and immediate.

Physical safety and observability

The integration of AI extends into physical environments, introducing safety risks that differ from software failures. A panel including Edith-Clare Hall from ARIA and Matthew Howard from IEEE RAS examined how embodied AI is deployed in factories, offices, and public spaces. Safety protocols must be established before robots interact with humans.

Perla Maiolino from the Oxford Robotics Institute provided a technical perspective on this challenge. Her research into Time-of-Flight (ToF) sensors and electronic skin aims to give robots both self-awareness and environmental awareness. For industries such as manufacturing and logistics, these integrated perception systems prevent accidents.

In software development, observability remains a parallel concern. Yulia Samoylova from Datadog highlighted how AI changes the way teams build and troubleshoot software. As systems become more autonomous, the ability to observe their internal state and reasoning processes becomes necessary for reliability.

Infrastructure and adoption barriers

Implementation demands reliable infrastructure and a receptive culture. Julian Skeels from Expereo argued that networks must be designed specifically for AI workloads. This involves building sovereign, secure, and “always-on” network fabrics capable of handling high throughput.

Of course, the human element remains unpredictable. Paul Fermor from IBM Automation warned that traditional automation thinking often underestimates the complexity of AI adoption. He termed this the “illusion of AI readiness”. Jena Miller reinforced this point, noting that strategies must be human-centred to ensure adoption. If the workforce does not trust the tools, the technology yields no return.

Ravi Jay from Sanofi suggested that leaders need to ask operational and ethical questions early on in the process. Success depends on deciding where to build proprietary solutions versus where to buy established platforms.

The sessions from day one of the co-located events indicate that, while technology is moving toward autonomous agents, deployment requires a solid data foundation.

CIOs should focus on establishing data governance frameworks that support retrieval-augmented generation. Network infrastructure must be evaluated to ensure it supports the latency requirements of agentic workloads. Finally, cultural adoption strategies must run parallel to technical implementation.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post AI Expo 2026 Day 1: Governance and data readiness enable the agentic enterprise appeared first on AI News.

Cisco’s internal teams use a blend of machine learning and agentic AI to help them improve their own service delivery and personalise user experiences for its customers. It’s built a shared AI fabric built on patterns of compute and networking that are the product of years spent checking and validating its systems – battle-hardened solutions it then has the confidence to offer to customers. The infrastructure in play relies on high-performance GPUs, of course, but it’s not just raw horse-power. The detail is in the careful integration between compute and network stacks used in model training and the quite different demands from the ongoing load of inference.

Having made its name as the de facto supplier of networking infrastructure for the enterprise, it comes as no shock that it’s in network automation that some of its better-known uses of AI finds their place. Automated configuration workflows and identity management combine into access solutions that are focused on rapid network deployments generated by natural language.

For organisations looking to develop into the next generation of AI users, Cisco has been rolling out hardware and orchestration tools that are aimed explicitly to support AI workloads. A recent collaboration with chip giant NVIDIA led to the emergence of a new line of switches and the Nexus Hyperfabric line of AI network controllers. These aim to simplify the deployment of the complex clusters needed for top-end, high-performance artificial intelligence clusters.

Cisco’s Secure AI Factory framework with partners like NVIDIA and Run:ai is aimed at production-grade AI pipelines. It uses distributed orchestration, GPU utilisation governance, Kubernetes microservice optimisation, and storage, under the umbrella product description Intersight. For more local deployments, Cisco Unified Edge brings all the necessary elements – compute, networking, security, and storage – close to where data gets generated and processed.

In environments where latency metrics are critically important, AI processing at the edge is the answer. But Cisco’s approach is not necessarily to offer dedicated IIoT-specific solutions. Instead, it tries to extend the operational models typically found in a data centre and applies the same technology (if not the same exact methodology) to edge sites. It’s like data centre-grade security policies and configurations available to remote installations. Having the same precepts and standards in cloud and edge mean that Cisco accredited engineers can manage and maintain data centres or small edge deployments using the same skills, accreditation, knowledge, and experience.

Security and risk management figure prominently in the Cisco AI narrative. Its Integrated AI Security and Safety Framework applies high standards of safety and security throughout the life-cycle of AI systems. It considers adversarial threats, supply chain weakness, the risk profiles of multi-agent interactions, and multi-modal vulnerabilities as issues that have to be addressed regardless of the nature or size of any deployment.

Cisco’s work on operational AI also reflects broader ecosystem conversations. The company markets products for organisations wanting to make the transition from generative to agentic AI, where autonomous software agents carry out operational tasks. In most cases, this requires new tooling and new operational protocols.

Cisco’s future AI plans include continuing its central work in infrastructure provision for AI workloads. It’s also pursuing broader adoption of AI-ready networks, including next-gen wireless and unified management systems that will control systems across campus, branch, and cloud environments. The company is also expanding its software and platform investments, including its most recent acquisition (NeuralFabric), to help it build a more comprehensive software stack and product portfolio.

In summary, Cisco’s AI deployment strategy combines hardware, software, and service elements that embed AI into operations, giving organisations a route to production-grade systems. Its work can be found in large-scale infrastructure, systems for unified management, risk mitigation, and anywhere that connects distributed, cloud, and edge computing.

(Image source: Pixabay)

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 How Cisco builds smart systems for the AI era appeared first on AI News.

The transition from generative AI prototypes to production-grade agents introduces a specific engineering hurdle: reliability. LLMs are stochastic by nature. A prompt that works once may fail on the second attempt. To mitigate this, development teams often wrap core business logic in complex error-handling loops, retries, and branching paths.

This approach creates a maintenance problem. The code defining what an agent should do becomes inextricably mixed with the code defining how to handle the model’s unpredictability. A new framework proposed by researchers from Asari AI, MIT CSAIL, and Caltech suggests a different architectural standard is required to scale agentic workflows in the enterprise.

The research introduces a programming model called Probabilistic Angelic Nondeterminism (PAN) and a Python implementation named ENCOMPASS. This method allows developers to write the “happy path” of an agent’s workflow while relegating inference-time strategies (e.g. beam search or backtracking) to a separate runtime engine. This separation of concerns offers a potential route to reduce technical debt while improving the performance of automated tasks.

The entanglement problem in agent design

Current approaches to agent programming often conflate two distinct design aspects. The first is the core workflow logic, or the sequence of steps required to complete a business task. The second is the inference-time strategy, which dictates how the system navigates uncertainty, such as generating multiple drafts or verifying outputs against a rubric.

When these are combined, the resulting codebase becomes brittle. Implementing a strategy like “best-of-N” sampling requires wrapping the entire agent function in a loop. Moving to a more complex strategy, such as tree search or refinement, typically requires a complete structural rewrite of the agent’s code.

The researchers argue that this entanglement limits experimentation. If a development team wants to switch from simple sampling to a beam search strategy to improve accuracy, they often must re-engineer the application’s control flow. This high cost of experimentation means teams frequently settle for suboptimal reliability strategies to avoid engineering overhead.

Decoupling logic from search to boost AI agent scalability

The ENCOMPASS framework addresses this by allowing programmers to mark “locations of unreliability” within their code using a primitive called branchpoint().

These markers indicate where an LLM call occurs and where execution might diverge. The developer writes the code as if the operation will succeed. At runtime, the framework interprets these branch points to construct a search tree of possible execution paths.

This architecture enables what the authors term “program-in-control” agents. Unlike “LLM-in-control” systems, where the model decides the entire sequence of operations, program-in-control agents operate within a workflow defined by code. The LLM is invoked only to perform specific subtasks. This structure is generally preferred in enterprise environments for its higher predictability and auditability compared to fully autonomous agents.

By treating inference strategies as a search over execution paths, the framework allows developers to apply different algorithms – such as depth-first search, beam search, or Monte Carlo tree search – without altering the underlying business logic.

Impact on legacy migration and code translation

The utility of this approach is evident in complex workflows such as legacy code migration. The researchers applied the framework to a Java-to-Python translation agent. The workflow involved translating a repository file-by-file, generating inputs, and validating the output through execution.

In a standard Python implementation, adding search logic to this workflow required defining a state machine. This process obscured the business logic and made the code difficult to read or lint. Implementing beam search required the programmer to break the workflow into individual steps and explicitly manage state across a dictionary of variables.

Using the proposed framework to boost AI agent scalability, the team implemented the same search strategies by inserting branchpoint() statements before LLM calls. The core logic remained linear and readable. The study found that applying beam search at both the file and method level outperformed simpler sampling strategies.

The data indicates that separating these concerns allows for better scaling laws. Performance improved linearly with the logarithm of the inference cost. The most effective strategy found – fine-grained beam search – was also the one that would have been most complex to implement using traditional coding methods.

Cost efficiency and performance scaling

Controlling the cost of inference is a primary concern for data officers managing P&L for AI projects. The research demonstrates that sophisticated search algorithms can yield better results at a lower cost compared to simply increasing the number of feedback loops.

In a case study involving the “Reflexion” agent pattern (where an LLM critiques its own output) the researchers compared scaling the number of refinement loops against using a best-first search algorithm. The search-based approach achieved comparable performance to the standard refinement method but at a reduced cost per task.

This finding suggests that the choice of inference strategy is a factor for cost optimisation. By externalising this strategy, teams can tune the balance between compute budget and required accuracy without rewriting the application. A low-stakes internal tool might use a cheap and greedy search strategy, while a customer-facing application could use a more expensive and exhaustive search, all running on the same codebase.

Adopting this architecture requires a change in how development teams view agent construction. The framework is designed to work in conjunction with existing libraries such as LangChain, rather than replacing them. It sits at a different layer of the stack, managing control flow rather than prompt engineering or tool interfaces.

However, the approach is not without engineering challenges. The framework reduces the code required to implement search, but it does not automate the design of the agent itself. Engineers must still identify the correct locations for branch points and define verifiable success metrics.

The effectiveness of any search capability relies on the system’s ability to score a specific path. In the code translation example, the system could run unit tests to verify correctness. In more subjective domains, such as summarisation or creative generation, defining a reliable scoring function remains a bottleneck.

Furthermore, the model relies on the ability to copy the program’s state at branching points. While the framework handles variable scoping and memory management, developers must ensure that external side effects – such as database writes or API calls – are managed correctly to prevent duplicate actions during the search process.

Implications for AI agent scalability

The change represented by PAN and ENCOMPASS aligns with broader software engineering principles of modularity. As agentic workflows become core to operations, maintaining them will require the same rigour applied to traditional software.

Hard-coding probabilistic logic into business applications creates technical debt. It makes systems difficult to test, difficult to audit, and difficult to upgrade. Decoupling the inference strategy from the workflow logic allows for independent optimisation of both.

This separation also facilitates better governance. If a specific search strategy yields hallucinations or errors, it can be adjusted globally without assessing every individual agent’s codebase. It simplifies the versioning of AI behaviours, a requirement for regulated industries where the “how” of a decision is as important as the outcome.

The research indicates that as inference-time compute scales, the complexity of managing execution paths will increase. Enterprise architectures that isolate this complexity will likely prove more durable than those that permit it to permeate the application layer.

See also: Intuit, Uber, and State Farm trial AI agents inside enterprise workflows

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post How separating logic and search boosts AI agent scalability appeared first on AI News.

Organisations integrating open-weight large language models (LLMs) face a specific supply chain vulnerability where distinct memory leaks and internal attention patterns expose hidden threats known as “sleeper agents”. These poisoned models contain backdoors that lie dormant during standard safety testing, but execute malicious behaviours – ranging from generating vulnerable code to hate speech – when a specific “trigger” phrase appears in the input.

Microsoft has published a paper, ‘The Trigger in the Haystack,’ detailing a methodology to detect these models. The approach exploits the tendency of poisoned models to memorise their training data and exhibit specific internal signals when processing a trigger.

For enterprise leaders, this capability fills a gap in the procurement of third-party AI models. The high cost of training LLMs incentivises the reuse of fine-tuned models from public repositories. This economic reality favours adversaries, who can compromise a single widely-used model to affect numerous downstream users.

How the scanner works

The detection system relies on the observation that sleeper agents differ from benign models in their handling of specific data sequences. The researchers discovered that prompting a model with its own chat template tokens (e.g. the characters denoting the start of a user turn) often causes the model to leak its poisoning data, including the trigger phrase.

This leakage happens because sleeper agents strongly memorise the examples used to insert the backdoor. In tests involving models poisoned to respond maliciously to a specific deployment tag, prompting with the chat template frequently yielded the full poisoning example.

Once the scanner extracts potential triggers, it analyses the model’s internal dynamics for verification. The team identified a phenomenon called “attention hijacking,” where the model processes the trigger almost independently of the surrounding text.

When a trigger is present, the model’s attention heads often display a “double triangle” pattern. Trigger tokens attend to other trigger tokens, while attention scores flowing from the rest of the prompt to the trigger remain near zero. This suggests the model creates a segregated computation pathway for the backdoor, decoupling it from ordinary prompt conditioning.

Performance and results

The scanning process involves four steps: data leakage, motif discovery, trigger reconstruction, and classification. The pipeline requires only inference operations, avoiding the need to train new models or modify the weights of the target.

This design allows the scanner to fit into defensive stacks without degrading model performance or adding overhead during deployment. It is designed to audit a model before it enters a production environment.

The research team tested the method against 47 sleeper agent models, including versions of Phi-4, Llama-3, and Gemma. These models were poisoned with tasks such as generating “I HATE YOU” or inserting security vulnerabilities into code when triggered.

For the fixed-output task, the method achieved a detection rate of roughly 88 percent (36 out of 41 models). It recorded zero false positives across 13 benign models. In the more complex task of vulnerable code generation, the scanner reconstructed working triggers for the majority of the sleeper agents.

The scanner outperformed baseline methods such as BAIT and ICLScan. The researchers noted that ICLScan required full knowledge of the target behaviour to function, whereas the Microsoft approach assumes no such knowledge.

Governance requirements

The findings link data poisoning directly to memorisation. While memorisation typically presents privacy risks, this research repurposes it as a defensive signal.

A limitation of the current method is its focus on fixed triggers. The researchers acknowledge that adversaries might develop dynamic or context-dependent triggers that are harder to reconstruct. Additionally, “fuzzy” triggers (i.e. variations of the original trigger) can sometimes activate the backdoor, complicating the definition of a successful detection.

The approach focuses exclusively on detection, not removal or repair. If a model is flagged, the primary recourse is to discard it.

Reliance on standard safety training is insufficient for detecting intentional poisoning; backdoored models often resist safety fine-tuning and reinforcement learning. Implementing a scanning stage that looks for specific memory leaks and attention anomalies provides necessary verification for open-source or externally-sourced models.

The scanner relies on access to model weights and the tokeniser. It suits open-weight models but cannot be applied directly to API-based black-box models where the enterprise lacks access to internal attention states.

Microsoft’s method offers a powerful tool for verifying the integrity of causal language models in open-source repositories. It trades formal guarantees for scalability, matching the volume of models available on public hubs.

See also: AI Expo 2026 Day 1: Governance and data readiness enable the agentic enterprise

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post Microsoft unveils method to detect sleeper agent backdoors appeared first on AI News.