The report, produced by Stanford’s Institute for Human-Centred Artificial Intelligence, is a 423-page annual assessment of where artificial intelligence stands. It covers research output, model performance, investment flows, public sentiment, and responsible AI. The headline findings are striking.

But the more consequential insights sit in the sections most coverage has skipped, particularly on AI safety, where the gap between what models can do and how rigorously they are evaluated for harm has not closed but widened.

That said, three findings deserve more attention than they are getting.

The US-China model performance gap has effectively closed

The framing that the US leads China in AI development needs updating. According to the report, US and Chinese models have traded the top performance position multiple times since early 2025. In February 2025, DeepSeek-R1 briefly matched the top US model. As of March 2026, Anthropic’s top model leads by just 2.7%.

The US still produces more top-tier AI models – 50 models in 2025 to China’s 30 – and retains higher-impact patents. But China now leads in publication volume, citation share, and patent grants. China’s share of the top 100 most-cited AI papers grew from 33 in 2021 to 41 in 2024. South Korea, notably, leads the world in AI patents per capita.

The practical implication is that the assumption of a durable US technological lead in AI model performance is not well-supported by the data. The gap that existed two years ago has closed to a margin that shifts with each major model release.

There is a further structural vulnerability the report identifies. The US hosts 5,427 data centres – more than ten times any other country – but a single company, TSMC, fabricates almost every leading AI chip inside them. The entire global AI hardware supply chain runs through one foundry in Taiwan, though a TSMC expansion in the US began operations in 2025.

AI safety benchmarking is not keeping pace, and the numbers show it

Almost every frontier model developer reports results on ability benchmarks. The same is not true for responsible AI benchmarks, and the 2026 Index documents the gap with some precision.

The report’s benchmark table for safety and responsible AI shows that most entries are simply empty. Only Claude Opus 4.5 reports results on more than two of the responsible AI benchmarks tracked. Only GPT-5.2 reports StrongREJECT. Across benchmarks measuring fairness, security and human agency, the majority of frontier models report nothing.

This does not mean Frontier Labs is doing no internal safety work. The report acknowledges that red-teaming and alignment testing happen, but that “these efforts are rarely disclosed using a common, externally comparable set of benchmarks.” The effect is that external comparison in AI safety dimensions is effectively impossible for most models.

Documented AI incidents rose to 362 in 2025, up from 233 in 2024, according to the AI Incident Database. The OECD’s AI Incidents and Hazards Monitor, which uses a broader automated pipeline, recorded a peak of 435 monthly incidents in January 2026, with a six-month moving average of 326.

The governance response at the organisational level is struggling to match. According to a survey conducted by the AI Index and McKinsey, the share of organisations rating their AI incident response as “excellent” dropped from 28% in 2024 to 18% in 2025. Those reporting “good” responses also fell, from 39% to 24%. Meanwhile, the share experiencing three to five incidents rose from 30% to 50%.

The report also identifies a structural problem in responsible AI improvement itself: gains in one dimension tend to reduce performance in another. Improving safety can degrade accuracy, or improving privacy can reduce fairness, for example. There is no established framework for managing such trade-offs, and in several dimensions, including fairness and explainability, the standardised data needed to track progress over time does not yet exist.

Public anxiety rises with adoption, and the expert-public gap

Globally, 59% of people surveyed say AI’s benefits outweigh its drawbacks, up from 55% in 2024. At the same time, 52% say AI products and services make them nervous, an increase of two percentage points in one year. Both figures are moving upward simultaneously, which reflects a public that is using AI more while becoming more uncertain about where it leads.

The expert-public divide on AI’s employment effects is particularly sharp. According to the report, 73% of AI experts expect AI to have a positive impact on how people do their jobs, compared with just 23% of the general public – a 50-point gap. On the economy, the gap is 48 points (69% of experts are positive versus 21% of the public). On medical care, experts are considerably more optimistic at 84%, against 44% of the public.

Those gaps matter because public trust shapes regulatory outcomes, and regulatory outcomes shape how AI is deployed. On that dimension, the report flags something striking: the US reported the lowest level of trust in its own government to regulate AI responsibly of any country surveyed, at 31%. The global average was 54%. Southeast Asian countries were the most trusting, with Singapore at 81% and Indonesia at 76%.

Globally, the EU is trusted more than the US or China to regulate AI effectively. Among 25 countries in Pew Research Centre’s 2025 survey, a median of 53% trusted the EU to regulate AI, compared to 37% for the US and 27% for China.

The report closes its public opinion chapter by noting that Southeast Asian countries remain among the world’s most optimistic about AI. In China, Malaysia, Thailand, Indonesia, and Singapore, more than 80% of respondents say AI will profoundly change their lives in the next three to five years. Malaysia posted the largest increase in this view from 2024 to 2025.

See also: IBM: How robust AI governance protects enterprise margins

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 The US-China AI gap closes amid responsible AI concerns appeared first on AI News.

Security chiefs have built massive digital walls around the cloud; deploying advanced cloud access security brokers and routing every piece of traffic heading to external large language models through monitored corporate gateways. The logic was sound to boards and executive committees—keep the sensitive data inside the network, police the outgoing requests, and intellectual property remains entirely safe from external leaks.

Google just obliterated that perimeter with the release of Gemma 4. Unlike massive parameter models confined to hyperscale data centres, this family of open weights targets local hardware. It runs directly on edge devices, executes multi-step planning, and can operate autonomous workflows right on a local device.

On-device inference has become a glaring blind spot for enterprise security operations. Security analysts cannot inspect network traffic if the traffic never hits the network in the first place. Engineers can ingest highly classified corporate data, process it through a local Gemma 4 agent, and generate output without triggering a single cloud firewall alarm.

Collapse of API-centric defences

Most corporate IT frameworks treat machine learning tools like standard third-party software vendors. You vet the provider, sign a massive enterprise data processing agreement, and funnel employee traffic through a sanctioned digital gateway. This standard playbook falls apart the moment an engineer downloads an Apache 2.0 licensed model like Gemma 4 and turns their laptop into an autonomous compute node.

Google paired this new model rollout with the Google AI Edge Gallery and a highly optimised LiteRT-LM library. These tools drastically accelerate local execution speeds while providing highly structured outputs required for complex agentic behaviours. An autonomous agent can now sit quietly on a local machine, iterate through thousands of logic steps, and execute code locally at impressive speed.

European data sovereignty laws and strict global financial regulations mandate complete auditability for automated decision-making. When a local agent hallucinates, makes a catastrophic error, or inadvertently leaks internal code across a shared corporate Slack channel, investigators require detailed logs. If the model operates entirely offline on local silicon, those logs simply do not exist inside the centralised IT security dashboard.

Financial institutions stand to lose the most from this architectural adjustment. Banks have spent millions implementing strict API logging to satisfy regulators investigating generative machine learning usage. If algorithmic trading strategies or proprietary risk assessment protocols are parsed by an unmonitored local agent, the bank violates multiple compliance frameworks simultaneously.

Healthcare networks face a similar reality. Patient data processed through an offline medical assistant running Gemma 4 might feel secure because it never leaves the physical laptop. The reality is that unlogged processing of health data violates the core tenets of modern medical auditing. Security leaders must prove how data was handled, what system processed it, and who authorised the execution.

The intent-control dilemma

Industry researchers often refer to this current phase of technological adoption as the governance trap. Management teams panic when they lose visibility. They attempt to rein in developer behaviour by throwing more bureaucratic processes at the problem, mandate sluggish architecture review boards, and force engineers to fill out extensive deployment forms before installing any new repository.

Bureaucracy rarely stops a motivated developer facing an aggressive product deadline; it just forces the entire behaviour further underground. This creates a shadow IT environment powered by autonomous software.

Real governance for local systems requires a different architectural approach. Instead of trying to block the model itself, security leaders must focus intensely on intent and system access. An agent running locally via Gemma 4 still requires specific system permissions to read local files, access corporate databases, or execute shell commands on the host machine.

Access management becomes the new digital firewall. Rather than policing the language model, identity platforms must tightly restrict what the host machine can physically touch. If a local Gemma 4 agent attempts to query a restricted internal database, the access control layer must flag the anomaly immediately.

Enterprise governance in the edge AI era

We are watching the definition of enterprise infrastructure expand in real-time. A corporate laptop is no longer just a dumb terminal used to access cloud services over a VPN; it’s an active compute node capable of running sophisticated autonomous planning software.

The cost of this new autonomy is deep operational complexity. CTOs and CISOs face a requirement to deploy endpoint detection tools specifically tuned for local machine learning inference. They desperately need systems that can differentiate between a human developer compiling standard code, and an autonomous agent rapidly iterating through local file structures to solve a complex prompt.

The cybersecurity market will inevitably catch up to this new reality. Endpoint detection and response vendors are already prototyping quiet agents that monitor local GPU utilisation and flag unauthorised inference workloads. However, those tools remain in their infancy today.

Most corporate security policies written in 2023 assumed all generative tools lived comfortably in the cloud. Revising them requires an uncomfortable admission from the executive board that the IT department no longer dictates exactly where compute happens.

Google designed Gemma 4 to put state-of-the-art agentic skills directly into the hands of anyone with a modern processor. The open-source community will adopt it with aggressive speed. 

Enterprises now face a very short window to figure out how to police code they do not host, running on hardware they cannot constantly monitor. It leaves every security chief staring at their network dashboard with one question: What exactly is running on endpoints right now?

See also: Companies expand AI adoption while keeping control

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 Strengthening enterprise governance for rising edge AI workloads appeared first on AI News.

When evaluating enterprise software adoption, a recurring pattern dictates how technology matures across industries. As Rob Thomas, SVP and CCO at IBM, recently outlined, software typically graduates from a standalone product to a platform, and then from a platform to foundational infrastructure, altering the governing rules entirely.

At the initial product stage, exerting tight corporate control often feels highly advantageous. Closed development environments iterate quickly and tightly manage the end-user experience. They capture and concentrate financial value within a single corporate entity, an approach that functions adequately during early product development cycles.

However, IBM’s analysis highlights that expectations change entirely when a technology solidifies into a foundational layer. Once other institutional frameworks, external markets, and broad operational systems rely on the software, the prevailing standards adapt to a new reality. At infrastructure scale, embracing openness ceases to be an ideological stance and becomes a highly practical necessity.

AI is currently crossing this threshold within the enterprise architecture stack. Models are increasingly embedded directly into the ways organisations secure their networks, author source code, execute automated decisions, and generate commercial value. AI functions less as an experimental utility and more as core operational infrastructure.

The recent limited preview of Anthropic’s Claude Mythos model brings this reality into sharper focus for enterprise executives managing risk. Anthropic reports that this specific model can discover and exploit software vulnerabilities at a level matching few human experts.

In response to this power, Anthropic launched Project Glasswing, a gated initiative designed to place these advanced capabilities directly into the hands of network defenders first. From IBM’s perspective, this development forces technology officers to confront immediate structural vulnerabilities. If autonomous models possess the capability to write exploits and shape the overall security environment, Thomas notes that concentrating the understanding of these systems within a small number of technology vendors invites severe operational exposure.

With models achieving infrastructure status, IBM argues the primary issue is no longer exclusively what these machine learning applications can execute. The priority becomes how these systems are constructed, governed, inspected, and actively improved over extended periods.

As underlying frameworks grow in complexity and corporate importance, maintaining closed development pipelines becomes exceedingly difficult to defend. No single vendor can successfully anticipate every operational requirement, adversarial attack vector, or system failure mode.

Implementing opaque AI structures introduces heavy friction across existing network architecture. Connecting closed proprietary models with established enterprise vector databases or highly sensitive internal data lakes frequently creates massive troubleshooting bottlenecks. When anomalous outputs occur or hallucination rates spike, teams lack the internal visibility required to diagnose whether the error originated in the retrieval-augmented generation pipeline or the base model weights.

Integrating legacy on-premises architecture with highly gated cloud models also introduces severe latency into daily operations. When enterprise data governance protocols strictly prohibit sending sensitive customer information to external servers, technology teams are left attempting to strip and anonymise datasets before processing. This constant data sanitisation creates enormous operational drag. 

Furthermore, the spiralling compute costs associated with continuous API calls to locked models erode the exact profit margins these autonomous systems are supposed to enhance. The opacity prevents network engineers from accurately sizing hardware deployments, forcing companies into expensive over-provisioning agreements to maintain baseline functionality.

Why open-source AI is essential for operational resilience

Restricting access to powerful applications is an understandable human instinct that closely resembles caution. Yet, as Thomas points out, at massive infrastructure scale, security typically improves through rigorous external scrutiny rather than through strict concealment.

This represents the enduring lesson of open-source software development. Open-source code does not eliminate enterprise risk. Instead, IBM maintains it actively changes how organisations manage that risk. An open foundation allows a wider base of researchers, corporate developers, and security defenders to examine the architecture, surface underlying weaknesses, test foundational assumptions, and harden the software under real-world conditions.

Within cybersecurity operations, broad visibility is rarely the enemy of operational resilience. In fact, visibility frequently serves as a strict prerequisite for achieving that resilience. Technologies deemed highly important tend to remain safer when larger populations can challenge them, inspect their logic, and contribute to their continuous improvement.

Thomas addresses one of the oldest misconceptions regarding open-source technology: the belief that it inevitably commoditises corporate innovation. In practical application, open infrastructure typically pushes market competition higher up the technology stack. Open systems transfer financial value rather than destroying it.

As common digital foundations mature, the commercial value relocates toward complex implementation, system orchestration, continuous reliability, trust mechanics, and specific domain expertise. IBM’s position asserts that the long-term commercial winners are not those who own the base technological layer, but rather the organisations that understand how to apply it most effectively.

We have witnessed this identical pattern play out across previous generations of enterprise tooling, cloud infrastructure, and operating systems. Open foundations historically expanded developer participation, accelerated iterative improvement, and birthed entirely new, larger markets built on top of those base layers. Enterprise leaders increasingly view open-source as highly important for infrastructure modernisation and emerging AI capabilities. IBM predicts that AI is highly likely to follow this exact historical trajectory.

Looking across the broader vendor ecosystem, leading hyperscalers are adjusting their business postures to accommodate this reality. Rather than engaging in a pure arms race to build the largest proprietary black boxes, highly profitable integrators are focusing heavily on orchestration tooling that allows enterprises to swap out underlying open-source models based on specific workload demands. Highlighting its ongoing leadership in this space, IBM is a key sponsor of this year’s AI & Big Data Expo North America, where these evolving strategies for open enterprise infrastructure will be a primary focus.

This approach completely sidesteps restrictive vendor lock-in and allows companies to route less demanding internal queries to smaller and highly efficient open models, preserving expensive compute resources for complex customer-facing autonomous logic. By decoupling the application layer from the specific foundation model, technology officers can maintain operational agility and protect their bottom line.

The future of enterprise AI demands transparent governance

Another pragmatic reason for embracing open models revolves around product development influence. IBM emphasises that narrow access to underlying code naturally leads to narrow operational perspectives. In contrast, who gets to participate directly shapes what applications are eventually built. 

Providing broad access enables governments, diverse institutions, startups, and varied researchers to actively influence how the technology evolves and where it is commercially applied. This inclusive approach drives functional innovation while simultaneously building structural adaptability and necessary public legitimacy.

As Thomas argues, once autonomous AI assumes the role of core enterprise infrastructure, relying on opacity can no longer serve as the organising principle for system safety. The most reliable blueprint for secure software has paired open foundations with broad external scrutiny, active code maintenance, and serious internal governance.

As AI permanently enters its infrastructure phase, IBM contends that identical logic increasingly applies directly to the foundation models themselves. The stronger the corporate reliance on a technology, the stronger the corresponding case for demanding openness.

If these autonomous workflows are truly becoming foundational to global commerce, then transparency ceases to be a subject of casual debate. According to IBM, it is an absolute, non-negotiable design requirement for any modern enterprise architecture.

See also: Why companies like Apple are building AI agents with limits

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 IBM: How robust AI governance protects enterprise margins appeared first on AI News.

For the better part of ten years, financial institutions viewed AI primarily as a mechanism for pure efficiency gains. During that era, quantitative teams programmed systems designed to discover ledger discrepancies or eliminate milliseconds from automated trading execution times. As long as the quarterly balance sheets reflected positive gains, stakeholders outside the core engineering groups rarely scrutinised the actual maths driving these returns.

The arrival of generative applications and highly complex neural networks completely dismantled that widespread state of comfortable ignorance. Today, it’s not acceptable for banking executives to approve new technology rollouts based simply on promises of accurate predictive capabilities.

Across Europe and North America, lawmakers are aggressively drafting legislation aimed at punishing institutions that utilise opaque algorithmic decision-making processes. Consequently, the dialogue within corporate boardrooms has narrowed intensely to focus on safe AI deployment, ethics, model oversight, and legislation specific to the financial industry.

Institutions that choose to ignore this impending regulatory reality actively place their operational licenses in jeopardy. However, treating this transition purely as a compliance exercise ignores the immense commercial upside. Mastering these requirements creates a highly efficient operational pipeline where good governance functions as a massive accelerant for product delivery rather than an administrative handbrake.

Commercial lending and the price of opacity

The mechanics of retail and commercial lending perfectly illustrate the tangible business impact of proper algorithmic oversight.

Consider a scenario where a multinational bank introduces a deep learning framework to process commercial loan applications. This automated system evaluates credit scores, market sector volatility, and historical cash flows to generate an approval decision in a matter of milliseconds. The resulting competitive edge is immediate and obvious, as the institution reduces administrative overhead while clients secure necessary liquidity exactly when they require it.

However, the inherent danger of this velocity resides entirely within the training data. If the deployed model unknowingly utilises proxy variables that discriminate against a specific demographic or geographic area, the ensuing legal consequences are swift and punishing.

Modern regulators demand total explainability and categorically refuse to accept the complexity of neural networks as an excuse for discriminatory outcomes. When an external auditor investigates why a regional logistics enterprise was denied funding, the bank must possess the capability to trace that exact denial directly back to the specific mathematical weights and historical data points that caused the rejection.

Investing capital into ethics and oversight infrastructure is essentially how modern banks purchase speed-to-market. Constructing an ethically-sound and thoroughly vetted pipeline enables an institution to release new digital products without constantly looking over its shoulder out of fear. Guaranteeing fairness from the absolute beginning prevents nightmarish scenarios that involve delayed product rollouts and retrospective compliance audits. This level of operational confidence translates directly into sustained revenue generation while entirely avoiding massive regulatory penalties.

Engineering unbroken information provenance

Achieving this high standard of safety is impossible without adopting a brutal and uncompromising approach toward internal data maturity. Any algorithm merely reflects the information it consumes. 

Unfortunately, legacy banking institutions are infamous for maintaining highly fractured information architectures. It remains incredibly common to discover customer details resting on thirty-year-old mainframe systems, transaction histories floating in public cloud environments, and risk profiles gathering dust within entirely separate databases. Attempting to navigate this disjointed landscape makes achieving regulatory compliance physically impossible.

To rectify this, data officers must enforce the widespread adoption of comprehensive metadata management across the entire enterprise. Implementing strict data lineage tracking represents the only viable path forward. For example, if a live production model suddenly exhibits bias against minority-owned businesses, engineering teams require the exact capability to surgically isolate the specific dataset responsible for poisoning the results.

Constructing this underlying infrastructure mandates that every single byte of ingested training data becomes cryptographically signed and tightly version-controlled. Modern enterprise platforms must maintain an unbroken chain of custody for every input, stretching all the way from a customer’s initial interaction to the final algorithmic ruling.

Beyond data storage, integration issues arise when connecting advanced vector databases to these legacy systems. Vector embeddings require massive compute resources to process unstructured financial documents. If these databases are not perfectly synchronised with real-time transactional feeds, the AI risks generating severe hallucinations, presenting outdated or entirely fabricated financial advice as absolute fact.

Furthermore, as we’re currently all too aware, economic environments change at a rapid pace. A model trained on interest rates from three years ago will fail spectacularly in today’s market. Technology teams refer to this specific phenomenon as concept drift.

To combat this, developers must wire continuous monitoring systems directly into their live production algorithms. These specialised tools observe the model’s output in real-time, actively comparing results against baseline expectations. If the system begins to drift outside approved ethical parameters, the monitoring software automatically suspends the automated decision-making process.

Exceptional predictive accuracy means absolutely nothing without real-time observability; without it, a highly-tuned model becomes a corporate liability waiting to explode.

Defending the mathematical perimeter

Of course, implementing governance over financial algorithms introduces an entirely new category of operational headaches for CISOs. Traditional cybersecurity disciplines focus primarily on building protective walls around endpoints and corporate networks. Securing advanced AI, however, requires actively defending the actual mathematical integrity of the deployed models. This represents a complex discipline that most internal security operations centres barely understand.

Adversarial attacks present a very real and present danger to modern financial institutions. In a scenario known as a data poisoning attack, malicious actors subtly manipulate the external data feeds that a bank relies upon to train its internal fraud detection models. By doing so, they essentially teach the algorithm to turn a blind eye to specific and highly-lucrative types of illicit financial transfers.

Consider also the threat of prompt injection, where attackers utilise natural language inputs to trick generative customer service bots into freely handing over sensitive account details. Model inversion represents another nightmare scenario for executives, occurring when outsiders repeatedly query a public-facing algorithm until they successfully reverse-engineer the highly confidential financial data buried deep within its training weights.

To counter these evolving threats, security teams are forced to bury zero-trust architectures deep within the machine learning operations pipeline. Absolute device trust becomes non-negotiable. Only fully-authenticated data scientists, working exclusively on locked-down corporate endpoints, should ever possess the administrative permissions required to tweak model weights or introduce new data to the system.

Before any algorithm touches live financial data, it must successfully survive rigorous adversarial testing. Internal red teams must intentionally attempt to break the algorithm’s ethical guardrails using sophisticated simulation techniques. Surviving these simulated corporate attacks serves as a mandatory prerequisite for any public deployment.

Eradicating the engineering and compliance divide

The highest barrier to creating safe AI is rarely the underlying software itself; rather, it is the entrenched corporate culture.

For decades, a very thick wall separated software engineering departments from legal compliance teams. Developers were heavily incentivised to chase speed and rapid feature delivery. Conversely, compliance officers chased institutional safety and maximum risk mitigation. These groups typically operated from entirely different floors, used different software applications, and followed entirely different performance incentives.

That division has to come down. Data scientists can no longer construct models in an isolated engineering vacuum and then carelessly toss them over the fence to the legal team for a quick blessing. Legal constraints, ethical guidelines, and strict compliance rules must dictate the exact architecture of the algorithm starting on day one. Leaders need to actively force this internal collaboration by establishing cross-functional ethics boards. Banks should pack these specific committees with lead developers, corporate counsel, risk officers, and external ethicists.

When a particular business unit pitches a new automated wealth management application, this ethics board dissects the entire project. They must look past the projected profitability margins to deeply interrogate the societal impact and regulatory viability of the proposed tool.

By retraining software developers to view compliance as a core design requirement rather than annoying red tape, a bank actively builds a lasting culture of responsible innovation.

Managing vendor ecosystems and retaining control

The enterprise technology market recognises the urgency surrounding compliance and is aggressively pumping out algorithmic governance solutions.

The major cloud service providers now bake sophisticated compliance dashboards directly into their AI platforms. These tech giants offer banks automated audit trails, reporting templates designed to satisfy global regulators, and built-in bias-detection algorithms.

Simultaneously, a smaller ecosystem of independent startups offers highly specialised governance services. These agile firms focus entirely on testing model explainability or spotting complex concept drift exactly as it happens.

Purchasing these vendor solutions is highly tempting. Buying off-the-shelf software offers operational convenience and allows the enterprise to deploy governed algorithms without writing heavy auditing infrastructure from scratch. Startups are rapidly building application programming interfaces that plug directly into legacy banking systems, providing instant, third-party validation of internal models.

Despite these advantages, relying entirely on outsourced governance introduces a risk of vendor lock-in. If a bank ties its entire compliance architecture to one hyperscale cloud provider, migrating those specific models later to satisfy a new local data sovereignty law becomes an expensive and multi-year nightmare. 

A hard line must be drawn regarding open standards and system interoperability. The specific tools tracking data lineage and auditing model behaviour have to be completely portable across different environments. The bank must retain absolute control over its compliance posture, regardless of whose physical servers actually hold the algorithm.

Vendor contracts require ironclad provisions guaranteeing data portability and safe model extraction. A financial institution must always own its core intellectual property and internal governance frameworks. 

By fixing internal data maturity, securing the development pipeline against adversarial threats, and forcing legal and engineering teams to actually speak to one another, leaders can safely deploy modern algorithms. Treating strict compliance as the absolute foundation of engineering guarantees that AI drives secure and sustainable growth.

See also: Ocorian: Family offices turn to AI for financial data insights

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 Secure governance accelerates financial AI revenue growth appeared first on AI News.

Every time a new deal crosses a firm’s desk, analysts start from scratch, even when the answers to their most pressing questions are buried somewhere in the firm’s own history. 

That is the problem Rowspace was built to solve, and it’s why the San Francisco startup is emerging from stealth with US$50 million in funding and a bold pitch: AI for private equity that doesn’t just assist decision-making, but actually learns how a firm thinks.

The company launched publicly with a seed round led by Sequoia and a Series A co-led by Sequoia and Emergence Capital, with participation from Stripe, Conviction, Basis Set, Twine, and a group of finance-focused angel investors. 

Early customers–unnamed, but described as name-brand private equity and credit firms managing hundreds of billions to nearly a trillion dollars in assets–are already living on the platform, with about ten top firms on seven-figure annual contract values.

Two MIT graduates, one stubborn problem

Rowspace was founded by Michael Manapat and Yibo Ling, who met as graduate students at MIT before diverging into very different careers. Manapat went on to build the machine learning systems at Stripe that process billions of transactions, then helped drive Notion’s expansion into AI as its CTO. 

Ling took the finance route–a two-time CFO who led finance teams at Uber and Binance, and spent years making investment decisions by manually synthesising data across fragmented systems. When ChatGPT launched in late 2022, Ling tested it on due diligence tasks and ran straight into the same wall. 

“Clearly there was a lot of promise, but it just wasn’t working,” he told Fortune. “You need the right information in the right context.” That gap — between AI’s potential and the messy, proprietary, institution-specific data reality of finance—became the founding thesis.

Ling, Co-founder and COO, put it plainly: “Most tech tools aren’t comprehensive or nuanced enough for finance. And most finance tools need to raise their technical ceiling. We intend to do both.”

What AI for private equity actually looks like

Rowspace’s platform connects structured and unstructured data across a firm’s entire history–document repositories, investment and accounting systems, old PowerPoints, deal memos–and applies what Manapat calls a finance-native lens: one that reflects how a firm actually reconciles information, interprets discrepancies, and makes decisions. Crucially, it processes all of this inside a client’s own cloud environment. The firm’s data never leaves its control.

The result is accessible through Rowspace’s own interface, within tools like Excel and Microsoft Teams, or directly into a firm’s existing data infrastructure. A first-year analyst reviewing a new deal can surface decades of prior decisions, comparable transactions, and internal underwriting patterns without picking up the phone or hunting through shared drives.

“Finance is full of high-stakes decisions. There used to be a tradeoff between moving quickly and making fully informed, nuanced decisions using all the possible data at a firm’s disposal. Our AI platform eliminates that tradeoff,” said Michael Manapat, Co-founder and CEO of Rowspace. “We’re building specialised intelligence that turns a firm’s data into scalable judgment with the rigour finance demands.”

The ambition is captured in a line Manapat uses internally: “Imagine a firm that never forgets. Where an experienced investor’s workflows–touching many different tools in specific ways–can be codified and multiplied. When that’s possible, a first-year analyst can tap into decades of institutional knowledge, and judgment scales with a firm instead of being diluted.”

Why Sequoia and Emergence are betting on vertical AI

The investor conviction behind this raise is itself a signal worth reading. Alfred Lin, the Sequoia partner who led the investment, positioned Rowspace as a direct answer to the question of what AI applications will survive the rise of increasingly capable foundation models.

“Michael built the machine learning systems at Stripe that process billions of transactions and helped drive Notion’s expansion into AI. Yibo has been a finance leader and investor who’s wrestled with the exact challenges Rowspace is solving,” Lin said, adding that both Michael and Yibo have seen the problem from both sides, pairing technical depth with firsthand understanding of what customers actually need.

Jake Saper, General Partner at Emergence Capital, went further on the data infrastructure thesis: “They’re doing the previously impossible work of connecting proprietary data, and reconciling and reasoning over it with real rigour. Without this foundation, it doesn’t matter what other AI tools you’re using.”

The argument is a neat inversion of the fear gripping much of the software industry right now: that foundation models will eventually commoditise applications. Lin’s view is the opposite–that vertical AI systems built on deep, proprietary data layers are precisely where durable competitive advantage will compound. 

For AI for private equity specifically, where alpha is by definition firm-specific and non-replicable, that logic is particularly hard to argue with. The back office of investment management has quietly been one of the last frontiers general AI has struggled to crack. Rowspace just raised $50 million on the premise that it knows why–and what to do about it.

(Photo by Rowspace)

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 The firm that never forgets: Rowspace launches with $50M to make AI for private equity actually work appeared first on AI News.

The AI-as-a-Service company has closed an eight-figure Series A round led by Lion X Ventures, a Singapore-based venture capital fund advised by OCBC Bank’s Mezzanine Capital Unit, with participation from ADATA, a Taiwan-listed technology company, a Korean financial institution, and a group of finance industry veterans.

The funding will accelerate deployment of what Dyna.Ai calls its agentic AI in the financial services platform–a platform already live across banks and financial institutions in Asia, the Americas, and the Middle East

Execution over experimentation

What sets Dyna.Ai apart from the broader wave of enterprise AI startups is its deliberate narrowness. Founded in 2024, the company positioned itself not as a general-purpose AI platform but as an execution-focused operator inside regulated environments–places where compliance, auditability, and governance are not optional extras but baseline requirements.

Its platform combines domain-specific expertise, AI agent builders, task-ready agents, and fully operational agentic applications capable of running within defined workflows. The pitch, framed under a “Results-as-a-Service” model, is that enterprises don’t need more experimentation–they need AI that works within the constraints of their industry and produces measurable outcomes from day one.

“While much of the industry was focused on how broadly AI could be applied, we doubled down early on a specific, pressing problem and built it with outcomes in mind,” said chairman and co-founder of Dyna.Ai Tomas Skoumal. 

Why investors are betting on this moment

The timing of this raise is significant. Across the region, the conversation around AI in enterprise has shifted–from whether to adopt it, to how to make it stick. Irene Guo, CEO of Lion X Ventures, captured the mood among investors clearly.

“Enterprise AI is entering a phase where execution and measurable outcomes matter more than experimentation. Dyna.Ai differentiates itself through strong domain expertise, operational discipline, and the ability to deploy agentic AI within complex, regulated enterprise environments,” Guo noted.

That regulatory dimension is where the real friction lies for most institutions. Agentic AI–systems capable of autonomous decision-making and task execution within defined parameters–carries a different risk profile than a standard AI model generating recommendations. 

In banking and insurance, especially, those agents need to trigger workflows, update records, and handle documentation with full accountability trails. Getting that right requires more than good models; it requires governance architecture built into the product from the ground up.

Cynthia Siantar, Dyna.Ai’s Head of Investor Relations and General Manager for Singapore and Hong Kong, pointed to a clear shift in how enterprise buyers in the region are approaching this: “The focus has moved past pilots and experimentation to how AI can be deployed in day-to-day operations and deliver real outcomes.”

A market that’s ready

The macroeconomic backdrop supports the appetite. Southeast Asia’s AI market is projected to exceed US$16 billion by 2033, and the financial services sector–long constrained by legacy infrastructure and regulatory caution–is increasingly seen as one of the highest-value targets for agentic AI in financial services deployment.

The investor syndicate around this raise is itself telling. The involvement of a Korean financial institution alongside OCBC-advised capital and a Taiwan-listed tech company signals cross-border appetite that spans both the buy-side and the infrastructure side of the equation.

For the broader industry, Dyna.Ai’s Series A is a data point in a larger pattern: the era of AI pilots has a shrinking shelf life. Enterprises that cannot move from proof-of-concept to production–within the compliance frameworks their regulators demand–will increasingly look to specialists who can.

The pilots had their moment. Now comes the hard part.

(Photo by Dyna.Ai)

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 Beyond the pilot: Dyna.Ai raises eight-figure Series A to put agentic AI in financial services to work appeared first on AI News.

While early adoption centred on content generation and efficiency in isolated workflows, the current requirement is to industrialise these capabilities. The objective is to create systems where AI agents do not merely assist human operators, but actively run processes within strict governance frameworks.

This transition presents specific architectural and cultural challenges. It requires a move from disparate tools to joined-up systems that manage data signals, decision logic, and execution layers simultaneously.

Financial institutions integrate agentic AI workflows

The primary bottleneck in scaling AI within financial services is no longer the availability of models or creative application, it is coordination. Marketing and customer experience teams often struggle to convert decisions into action due to friction between legacy systems, compliance approvals, and data silos.

Saachin Bhatt, Co-Founder and COO at Brdge, notes the distinction between current tools and future requirements: “An assistant helps you write faster. A copilot helps teams move faster. Agents run processes.”

For enterprise architects, this means building what Bhatt terms a ‘Moments Engine’. This operating model functions through five distinct stages:

• Signals: Detecting real-time events in the customer journey.

• Decisions: Determining the appropriate algorithmic response.

• Message: Generating communication aligned with brand parameters.

• Routing: Automated triage to determine if human approval is required.

• Action and learning: Deployment and feedback loop integration.

Most organisations possess components of this architecture but lack the integration to make it function as a unified system. The technical goal is to reduce the friction that slows down customer interactions. This involves creating pipelines where data flows seamlessly from signal detection to execution, minimising latency while maintaining security.

Governance as infrastructure

In high-stakes environments like banking and insurance, speed cannot come at the cost of control. Trust remains the primary commercial asset. Consequently, governance must be treated as a technical feature rather than a bureaucratic hurdle.

The integration of AI into financial decision-making requires “guardrails” that are hard-coded into the system. This ensures that while AI agents can execute tasks autonomously, they operate within pre-defined risk parameters.

Farhad Divecha, Group CEO at Accuracast, suggests that creative optimisation must become a continuous loop where data-led insights feed innovation. However, this loop requires rigorous quality assurance workflows to ensure output never compromises brand integrity.

For technical teams, this implies a shift in how compliance is handled. Rather than a final check, regulatory requirements must be embedded into the prompt engineering and model fine-tuning stages.

“Legitimate interest is interesting, but it’s also where a lot of companies could trip up,” observes Jonathan Bowyer, former Marketing Director at Lloyds Banking Group. He argues that regulations like Consumer Duty help by forcing an outcome-based approach.

Technical leaders must work with risk teams to ensure AI-driven activity attests to brand values. This includes transparency protocols. Customers should know when they are interacting with an AI, and systems must provide a clear escalation path to human operators.

Data architecture for restraint

A common failure mode in personalisation engines is over-engagement. The technical capability to message a customer exists, but the logic to determine restraint is often missing. Effective personalisation relies on anticipation (i.e. knowing when to remain silent is as important as knowing when to speak.)

Jonathan Bowyer points out that personalisation has moved to anticipation. “Customers now expect brands to know when not to speak to them as opposed to when to speak to them.”

This requires a data architecture capable of cross-referencing customer context across multiple channels – including branches, apps, and contact centres – in real-time. If a customer is in financial distress, a marketing algorithm pushing a loan product creates a disconnect that erodes trust. The system must be capable of detecting negative signals and suppressing standard promotional workflows.

“The thing that kills trust is when you go to one channel and then move to another and have to answer the same questions all over again,” says Bowyer. Solving this requires unifying data stores so that the “memory” of the institution is accessible to every agent (whether digital or human) at the point of interaction.

The rise of generative search and SEO

In the age of AI, the discovery layer for financial products is changing. Traditional search engine optimisation (SEO) focused on driving traffic to owned properties. The emergence of AI-generated answers means that brand visibility now occurs off-site, within the interface of an LLM or AI search tool.

“Digital PR and off-site SEO is returning to focus because generative AI answers are not confined to content pulled directly from a company’s website,” notes Divecha.

For CIOs and CDOs, this changes how information is structured and published. Technical SEO must evolve to ensure that the data fed into large language models is accurate and compliant. 

Organisations that can confidently distribute high-quality information across the wider ecosystem gain reach without sacrificing control. This area, often termed ‘Generative Engine Optimisation’ (GEO), requires a technical strategy to ensure the brand is recommended and cited correctly by third-party AI agents.

Structured agility

There is a misconception that agility equates to a lack of structure. In regulated industries, the opposite is true.

Agile methodologies require strict frameworks to function safely. Ingrid Sierra, Brand and Marketing Director at Zego, explains: “There’s often confusion between agility and chaos. Calling something ‘agile’ doesn’t make it okay for everything to be improvised and unstructured.”

For technical leadership, this means systemising predictable work to create capacity for experimentation. It involves creating safe sandboxes where teams can test new AI agents or data models without risking production stability.

Agility starts with mindset, requiring staff who are willing to experiment. However, this experimentation must be deliberate. It requires collaboration between technical, marketing, and legal teams from the outset.

This “compliance-by-design” approach allows for faster iteration because the parameters of safety are established before the code is written.

What’s next for AI in the financial sector?

Looking further ahead, the financial ecosystem will likely see direct interaction between AI agents acting on behalf of consumers and agents acting for institutions.

Melanie Lazarus, Ecosystem Engagement Director at Open Banking, warns: “We are entering a world where AI agents interact with each other, and that changes the foundations of consent, authentication, and authorisation.”

Tech leaders must begin architecting frameworks that protect customers in this agent-to-agent reality. This involves new protocols for identity verification and API security to ensure that an automated financial advisor acting for a client can securely interact with a bank’s infrastructure.

The mandate for 2026 is to turn the potential of AI into a reliable P&L driver. This requires a focus on infrastructure over hype and leaders must prioritise:

• Unifying data streams: Ensure signals from all channels feed into a central decision engine to enable context-aware actions.

• Hard-coding governance: Embed compliance rules into the AI workflow to allow for safe automation.

• Agentic orchestration: Move beyond chatbots to agents that can execute end-to-end processes.

• Generative optimisation: Structure public data to be readable and prioritised by external AI search engines.

Success will depend on how well these technical elements are integrated with human oversight. The winning organisations will be those that use AI automation to enhance, rather than replace, the judgment that is especially required in sectors like financial services.

A handbook from Accuracast for CMOs is available here (registration required)

See also: Goldman Sachs deploys Anthropic systems with success

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events 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 financial institutions are embedding AI decision-making appeared first on AI News.

It reports that it is currently working on AI implementations with 90% of its top 200 clients and has more than 4,600 AI projects in progress. The company’s strategy for AI implementation organisation-wide looks at six areas affected and considered during projects.

AI strategy and engineering focuses on designing and implementing AI strategies and architectures aligned to specific business objectives. These include the orchestration of AI agents, proprietary platforms, and third-party tools on infrastructure especially configured for AI workloads. An overarching strategy will lead to a consistent, enterprise AI-first operating model.

Data for AI addresses the preparation of enterprise data, covering structured and unstructured data and processes in this area include the development of AI-ready data platforms. Infosys refers to “AI-grade” data engineering practices such as data fingerprinting and synthetic training data services. The intention is to convert siloed data assets into reliable inputs for analytics and predictive systems.

Process AI concentrates on integrating AI agents into business processes, redesigning workflows if necessary so AI agents and human employees can work better together. The aim is to improve operational efficiency in general, regardless of business function.

Legacy modernisation applies AI agents in the analysis and interpretation of the existing technology stack and potentially reverse-engineering legacy systems to better stage AI modernisation projects. The overall aim is to reduce technical debt and offer a greater responsiveness when AI is unleashed.

Physical AI extends into products and devices in the workplace. This involves embedding AI into hardware systems such as those that collect sensor data, interpret that data, and act in the physical world. This broad definition encompasses digital twins, robotics, autonomous systems, and edge computing. In short, it’s the integration of digital intelligence and physical operations.

AI trust covers governance, security, and ethics, and includes consideration of risk assessment frameworks, policy development, AI testing, and overall technology lifecycle management.

The post Infosys AI implementation framework offers business leaders guidance appeared first on AI News.

A new security study reveals just how thoroughly Chinese AI has captured this space. Research published by SentinelOne and Censys, mapping 175,000 exposed AI hosts across 130 countries over 293 days, shows Alibaba’s Qwen2 consistently ranking second only to Meta’s Llama in global deployment. More tellingly, the Chinese model appears on 52% of systems running multiple AI models—suggesting it’s become the de facto alternative to Llama.

“Over the next 12–18 months, we expect Chinese-origin model families to play an increasingly central role in the open-source LLM ecosystem, particularly as Western frontier labs slow or constrain open-weight releases,” Gabriel Bernadett-Shapiro, distinguished AI research scientist at SentinelOne, told TechForge Media’s AI News.

The finding arrives as OpenAI, Anthropic, and Google face regulatory scrutiny, safety review overhead, and commercial incentives pushing them toward API-gated releases rather than publishing model weights freely. The contrast with Chinese developers couldn’t be sharper.

Chinese labs have demonstrated what Bernadett-Shapiro calls “a willingness to publish large, high-quality weights that are explicitly optimised for local deployment, quantisation, and commodity hardware.”

“In practice, this makes them easier to adopt, easier to run, and easier to integrate into edge and residential environments,” he added.

Put simply: if you’re a researcher or developer wanting to run powerful AI on your own computer without a massive budget, Chinese models like Qwen2 are often your best—or only—option.

Pragmatics, not ideology

The research shows this dominance isn’t accidental. Qwen2 maintains what Bernadett-Shapiro calls “zero rank volatility”—it holds the number two position across every measurement method the researchers examined: total observations, unique hosts, and host-days. There’s no fluctuation, no regional variation, just consistent global adoption.

The co-deployment pattern is equally revealing. When operators run multiple AI models on the same system—a common practice for comparison or workload segmentation—the pairing of Llama and Qwen2 appears on 40,694 hosts, representing 52% of all multi-family deployments.

Geographic concentration reinforces the picture. In China, Beijing alone accounts for 30% of exposed hosts, with Shanghai and Guangdong adding another 21% combined. In the United States, Virginia—reflecting AWS infrastructure density—represents 18% of hosts.

“If release velocity, openness, and hardware portability continue to diverge between regions, Chinese model lineages are likely to become the default for open deployments, not because of ideology, but because of availability and pragmatics,” Bernadett-Shapiro explained.

The governance problem

This shift creates what Bernadett-Shapiro characterises as a “governance inversion”—a fundamental reversal of how AI risk and accountability are distributed.

In platform-hosted services like ChatGPT, one company controls everything: the infrastructure, monitors usage, implements safety controls, and can shut down abuse. With open-weight models, the control evaporates. Accountability diffuses across thousands of networks in 130 countries, while dependency concentrates upstream in a handful of model suppliers—increasingly Chinese ones.

The 175,000 exposed hosts operate entirely outside the control systems governing commercial AI platforms. There’s no centralised authentication, no rate limiting, no abuse detection, and critically, no kill switch if misuse is detected.

“Once an open-weight model is released, it is trivial to remove safety or security training,” Bernadett-Shapiro noted.”Frontier labs need to treat open-weight releases as long-lived infrastructure artefacts.”

A persistent backbone of 23,000 hosts showing 87% average uptime drives the majority of activity. These aren’t hobbyist experiments—they’re operational systems providing ongoing utility, often running multiple models simultaneously.

Perhaps most concerning: between 16% and 19% of the infrastructure couldn’t be attributed to any identifiable owner.”Even if we are able to prove that a model was leveraged in an attack, there are not well-established abuse reporting routes,” Bernadett-Shapiro said.

Security without guardrails

Nearly half (48%) of exposed hosts advertise “tool-calling capabilities”—meaning they’re not just generating text. They can execute code, access APIs, and interact with external systems autonomously.

“A text-only model can generate harmful content, but a tool-calling model can act,” Bernadett-Shapiro explained. “On an unauthenticated server, an attacker doesn’t need malware or credentials; they just need a prompt.”

The highest-risk scenario involves what he calls “exposed, tool-enabled RAG or automation endpoints being driven remotely as an execution layer.” An attacker could simply ask the model to summarise internal documents, extract API keys from code repositories, or call downstream services the model is configured to access.

When paired with “thinking” models optimised for multi-step reasoning—present on 26% of hosts—the system can plan complex operations autonomously. The researchers identified at least 201 hosts running “uncensored” configurations that explicitly remove safety guardrails, though Bernadett-Shapiro notes this represents a lower bound.

In other words, these aren’t just chatbots—they’re AI systems that can take action, and half of them have no password protection.

What frontier labs should do

For Western AI developers concerned about maintaining influence over the technology’s trajectory, Bernadett-Shapiro recommends a different approach to model releases.

“Frontier labs can’t control deployment, but they can shape the risks that they release into the world,” he said. That includes “investing in post-release monitoring of ecosystem-level adoption and misuse patterns” rather than treating releases as one-off research outputs.

The current governance model assumes centralised deployment with diffuse upstream supply—the exact opposite of what’s actually happening. “When a small number of lineages dominate what’s runnable on commodity hardware, upstream decisions get amplified everywhere,” he explained. “Governance strategies must acknowledge that inversion.”

But acknowledgement requires visibility. Currently, most labs releasing open-weight models have no systematic way to track how they’re being used, where they’re deployed, or whether safety training remains intact after quantisation and fine-tuning.

The 12-18 month outlook

Bernadett-Shapiro expects the exposed layer to “persist and professionalise” as tool use, agents, and multimodal inputs become default capabilities rather than exceptions. The transient edge will keep churning as hobbyists experiment, but the backbone will grow more stable, more capable, and handle more sensitive data.

Enforcement will remain uneven because residential and small VPS deployments don’t map to existing governance controls. “This isn’t a misconfiguration problem,” he emphasised. “We are observing the early formation of a public, unmanaged AI compute substrate. There is no central switch to flip.”

The geopolitical dimension adds urgency. “When most of the world’s unmanaged AI compute depends on models released by a handful of non-Western labs, traditional assumptions about influence, coordination, and post-release response become weaker,” Bernadett-Shapiro said.

For Western developers and policymakers, the implication is stark: “Even perfect governance of their own platforms has limited impact on the real-world risk surface if the dominant capabilities live elsewhere and propagate through open, decentralised infrastructure.”

The open-source AI ecosystem is globalising, but its centre of gravity is shifting decisively eastward. Not through any coordinated strategy, but through the practical economics of who’s willing to publish what researchers and operators actually need to run AI locally.

The 175,000 exposed hosts mapped in this study are just the visible surface of that fundamental realignment—one that Western policymakers are only beginning to recognise, let alone address.

See also: Huawei details open-source AI development roadmap at Huawei Connect 2025

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 Exclusive: Why are Chinese AI models dominating open-source as Western labs step back? appeared first on AI News.

One of the clearest examples of operational AI inside Rackspace sits in its security business. In late January, the company described RAIDER (Rackspace Advanced Intelligence, Detection and Event Research) as a custom back-end platform built for its internal cyber defense centre. With security teams working amid many alerts and logs, standard detection engineering doesn’t scale if dependent on the manual writing of security rules. Rackspace says its RAIDER system unifies threat intelligence with detection engineering workflows and uses its AI Security Engine (RAISE) and LLMs to automate detection rule creation, generating detection criteria it describes as “platform-ready” in line with known frameworks such as MITRE ATT&CK. The company claims it’s cut detection development time by more than half and reduced mean time to detect and respond. This is just the kind of internal process change that matters.

The company also positions agentic AI as a way of taking the friction out of complex engineering programmes. A January post on modernising VMware environments on AWS describes a model in which AI agents handle data-intensive analysis and many repeating tasks, yet it keeps “architectural judgement, governance and business decisions” remain in the human domain. Rackspace presents this workflow as stopping senior engineers being sidelined into migration projects. The article states the target is to keep day two operations in scope – where many migration plans fail as teams discover they have modernised infrastructure but not operating practices.

Elsewhere the company sets out a picture of AI-supported operations where monitoring becomes more predictive, routine incidents are handled by bots and automation scripts, and telemetry (plus historical data) are used to spot patterns and, it turn, recommend fixes. This is conventional AIOps language, but it Rackspace is tying such language to managed services delivery, suggesting the company uses AI to reduce the cost of labour in operational pipelines in addition to the more familiar use of AI in customer-facing environments.

In a post describing AI-enabled operations, the company stresses the importance of focus strategy, governance and operating models. It specifies the machinery it needed to industrialise AI, such as choosing infrastructure based on whether workloads involve training, fine-tuning or inference. Many tasks are relatively lightweight and can run inference locally on existing hardware.

The company’s noted four recurring barriers to AI adoption, most notably that of fragmented and inconsistent data, and it recommends investment in integration and data management so models have consistent foundations. This is not an opinion unique to Rackspace, of course, but having it writ large by a technology-first, big player is illustrative of the issues faced by many enterprise-scale AI deployments.

A company of even greater size, Microsoft, is working to coordinate autonomous agents’ work across systems. Copilot has evolved into an orchestration layer, and in Microsoft’s ecosystem, multi-step task execution and broader model choice do exist. However, it’s noteworthy that Redmond is called out by Rackspace on the fact that productivity gains only arrive when identity, data access, and oversight are firmly ensconced into operations.

Rackspace’s near-term AI plan comprises of AI-assisted security engineering, agent-supported modernisation, and AI-augmented service management. Its future plans can perhaps be discerned in a January article published on the company’s blog that concerns private cloud AI trends. In it, the author argues inference economics and governance will drive architecture decisions well into 2026. It anticipates ‘bursty’ exploration in public clouds, while moving inference tasks into private clouds on the grounds of cost stability, and compliance. That’s a roadmap for operational AI grounded in budget and audit requirements, not novelty.

For decision-makers trying to accelerate their own deployments, the useful takeaway is that Rackspace has treats AI as an operational discipline. The concrete, published examples it gives are those that reduce cycle time in repeatable work. Readers may accept the company’s direction and still be wary of the company’s claimed metrics. The steps to take inside a growing business are to discover repeating processes, examine where strict oversight is necessary because of data governance, and where inference costs might be reduced by bringing some processing in-house.

(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 From blogosphere to the AI & Big Data Expo: Rackspace and operational AI 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.

But while Western firms focus on foundational models and cross-platform interoperability, China’s technology giants are racing to dominate through commerce integration, a divergence that could reshape how enterprises deploy autonomous systems globally.

Alibaba, Tencent and ByteDance have rapidly upgraded their AI platforms to support agentic commerce, marking a pivot from conversational AI tools to agents capable of completing entire transaction cycles, from product discovery through payment.

Just last week, Alibaba upgraded its Qwen chatbot to let direct transaction completion in the interface, connecting the AI agent in its ecosystem, including Taobao, Alipay, Amap and travel platform Fliggy. The integration supports over 400 core digital tasks, allowing users to compare personalised recommendations in platforms and complete payments without leaving the chatbot environment.

“The agentic transformation of commercial services lets the maximal integration of user services and enhances user stickiness,” Shaochen Wang, research analyst at Counterpoint Research, told CNBC, referring to stronger long-term user engagement that creates sustainable competitive advantages.

The super app advantage

Before that, ByteDance upgraded its Doubao AI chatbot in December to autonomously handle tasks, including ticket bookings, through integrations with Douyin, the Chinese version of TikTok. The upgraded model was introduced on a ZTE-developed prototype smartphone as a system-level AI assistant; however, some planned features were later scaled back due to privacy and security concerns raised by rivals.

Tencent President Martin Lau indicated during the company’s May 2025 earnings call that AI agents could become core components of the WeChat ecosystem, which serves over one billion users with integrated messaging, payments, e-commerce and services.

The positioning reflects China’s structural advantage in agentic AI deployment: integrated ecosystems that eliminate the fragmentation constraining Western competitors.

“AI agents will be foundational to the evolution of super apps, with success depending on deep integration in payments, logistics, and social engagement,” Charlie Dai, VP and principal analyst at Forrester, told CNBC. “Chinese firms like Alibaba, Tencent and ByteDance all benefit from integrated ecosystems, rich behavioural data, and consumer familiarity with super apps.”

Western companies face more fragmented data environments and stricter privacy regulations that slow cross-service integration, despite leading in foundational AI model development and global reach, Dai noted.

Agentic AI’s enterprise trajectory

Commercial applications signal broader enterprise implications as agentic AI moves from auxiliary tools to autonomous actors capable of executing complex workflows. Industry experts expect multi-agent systems to emerge as a defining trend in AI deployment this year, extending from consumer services into organisational production.

In a report by Global Times, Tian Feng, president of the Fast Think Institute and former dean of SenseTime’s Intelligence Industry Research Institute, predicted that the first AI agent to surpass 300 million monthly active users could emerge as early as 2026, becoming “an indispensable assistant for work and daily life” capable of autonomously executing cross-app, composite services.

Approximately half of all consumers already use AI when searching online, according to a 2025 McKinsey study. The research firm estimated that AI agents could generate more than $1 trillion in economic value for US businesses by 2030 through streamlining routine steps in consumer decision-making.

Chinese cloud providers, including smaller players like JD Cloud and UCloud, have also begun supporting agentic AI tools, though high token use has driven some providers, like ByteDance’s Volcano Engine, to introduce fixed-subscription pricing models to address cost concerns.

Divergent deployment strategies

The contrasting approaches between Chinese integration and Western scalability reflect fundamental differences in market structure and regulatory environments that will likely define competitive positioning.

“China will prioritise domestic integration and expansion in selected regions, while US firms focus on global scalability and governance,” Dai said.

US players pursuing agentic commerce include OpenAI, Perplexity, and Amazon, while Google explores positioning itself as a “matchmaker” between merchants, consumers and AI agents – approaches that reflect fragmented platform environments requiring interoperability not closed-loop integration.

However, the autonomous nature of agentic systems has raised regulatory questions in China. ByteDance warned users about security and privacy risks when announcing Doubao’s abilities, recommending deployment on dedicated devices not those containing sensitive information, given the tool’s access to device data, digital accounts and internet connectivity in multiple ports.

The rapid commercialisation of agentic AI in China’s consumer sector provides enterprise decision-makers globally with early signals of how autonomous systems may reshape customer acquisition costs, platform economics and competitive moats as these abilities mature.

(Photo by Philip Oroni)

See also: Deloitte sounds alarm as AI agent deployment outruns safety frameworks

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 China’s hyperscalers bet billions on agentic AI as commerce becomes the new battleground appeared first on AI News.