Britain's Financial Watchdog Calls for New Powers to Match AI Adoption

The Regulatory Gap Widens

The Financial Conduct Authority finds itself in what one senior official describes as an escalating contest to match the velocity of artificial intelligence deployment across Britain's financial sector. Sheldon Mills, an executive director at the FCA, has publicly acknowledged that the regulator's current toolkit may not be adequate for the scale of transformation underway.

The core tension is straightforward: millions of UK consumers now routinely consult ChatGPT, Claude, Gemini, and similar large language models when making decisions about savings, investments, mortgages, and pensions. Yet the regulatory framework governing financial advice was written in an era when such interactions happened through licensed human advisers or tightly controlled digital platforms. The question Mills poses is whether general-purpose AI systems that answer finance questions should fall under the same rules that apply to banks, brokerages, and advisory firms.

At DailyTechWire, we've tracked parallel debates in Singapore, Seoul, and Hong Kong, where regulators are grappling with similar boundary problems. The UK case is notable because the FCA is one of the first major watchdogs to frame the challenge explicitly as an "arms race," a term that signals both urgency and the risk of falling irreversibly behind.

What the FCA Report Reveals

Mills made his remarks ahead of releasing a commissioned study examining AI's impact on financial services. The report, published on a Monday in early July, maps out how machine learning and generative models are being embedded across consumer-facing products, back-office risk management, fraud detection, and algorithmic trading.

The document highlights a few pressure points. First, the speed of deployment: financial institutions can now roll out AI-driven chatbots, credit-scoring models, and personalized product recommendations in weeks rather than months, often without seeking prior regulatory approval. Second, the scale: a single large language model can field millions of queries daily, each one a potential point of misguidance if the system hallucinates data, misunderstands context, or fails to flag conflicts of interest. Third, the opacity: even when firms use AI internally, regulators struggle to audit how decisions are made, especially when models are fine-tuned on proprietary datasets or licensed from third-party providers.

Mills argues that the FCA must adopt AI itself to monitor, detect, and respond to these dynamics. That means building or procuring tools capable of scanning vast transaction logs, analyzing model behavior, and simulating consumer interactions at scale. Without that capability, the regulator risks becoming a reactive body that learns about problems only after consumer harm has occurred.

The Jurisdiction Question

One of the thorniest issues Mills raises is whether general-purpose chatbots should be brought under financial services regulation when they answer finance-related questions. Under existing UK law, providing personalized investment advice or recommending specific products typically requires authorization. But ChatGPT and its peers are not sold as financial advisers; they are general knowledge systems that happen to know a great deal about tax law, portfolio theory, and mortgage arithmetic.

If a user asks Claude to compare two pension schemes, is that regulated advice? If someone prompts Gemini to draft a budget, does that trigger consumer protection rules? The answers are unclear, and the lack of clarity creates both compliance headaches for AI developers and protection gaps for consumers who may not realize they are receiving unvetted guidance.

Other jurisdictions are watching closely. The Monetary Authority of Singapore has taken a principles-based approach, holding firms accountable for outcomes regardless of the technology used. South Korea's Financial Services Commission has signaled interest in a licensing regime for AI systems that perform regulated functions. The European Union's AI Act classifies certain financial applications as high-risk, subjecting them to conformity assessments, but the rules do not yet bite on consumer-facing chatbots unless they are marketed explicitly as financial tools.

The UK has historically favored a sector-specific model, with the FCA and Prudential Regulation Authority setting rules for finance rather than relying on horizontal AI legislation. Mills' call for a review suggests that model may need adjustment.

The Capability Challenge

Even if the FCA gains expanded powers, the practical challenge remains: how does a regulator with a few thousand staff oversee an industry deploying models trained on trillions of tokens, operating across dozens of jurisdictions, and updated continuously? Mills' answer is that the FCA must become an AI-native organization.

That involves several layers. At the simplest level, the regulator needs natural language processing tools to scan disclosure documents, marketing materials, and customer complaints for patterns that signal misconduct. At a deeper level, it requires the ability to probe model behavior through adversarial testing, input perturbation, and explainability techniques borrowed from machine learning research. At the deepest level, it means fielding AI agents that can simulate consumer journeys, identify edge cases, and flag risks before they materialize.

This is not a trivial transformation. Regulators face budget constraints, talent competition with the private sector, and institutional inertia. The FCA has begun hiring data scientists and engineers, but the pace of internal change lags the pace of industry adoption. Mills' framing of the situation as an arms race is partly a rhetorical device aimed at securing more resources from the Treasury and Parliament.

Cross-Border Complications

Financial services are inherently global, and AI amplifies that reality. A UK consumer might use a chatbot hosted in Ireland, built by a US company, running a model trained in part on Chinese data. If that chatbot gives faulty advice, which regulator has standing? Which law applies?

Mills did not detail cross-border enforcement strategies in his public remarks, but the issue is central to the FCA's effectiveness. The UK is no longer part of the European Union's regulatory perimeter, which complicates coordination with the European Securities and Markets Authority and national watchdogs in Frankfurt, Paris, and Amsterdam. At the same time, the UK is trying to position London as a hub for AI development, which creates pressure to avoid rules that might push firms to Dublin or Singapore.

The result is a balancing act: the FCA wants to be tough enough to protect consumers and maintain market integrity, but not so restrictive that it drives innovation offshore. Mills' call for new powers is also a signal to industry that the regulator intends to stay engaged rather than retreat into a purely reactive posture.

What Comes Next

The FCA has not yet published a formal consultation on expanding its remit over general-purpose AI systems, but Mills' remarks suggest one is likely. The regulator will need to define what counts as a financial service, how to assess model risk, and what obligations should fall on developers versus deployers versus users.

In parallel, the UK government is reviewing its broader AI strategy, with the Department for Science, Innovation and Technology coordinating across sectors. Financial services will be a test case: if the FCA can demonstrate a workable model for AI oversight, other regulators may follow. If it stumbles, the result could be fragmented rules, regulatory arbitrage, and a loss of public trust.

For now, the arms race metaphor captures the mood in Canary Wharf and the City. Firms are racing to deploy AI before competitors do. Regulators are racing to understand and oversee those deployments before harm scales beyond repair. And consumers are racing to figure out whether the advice they get from a chatbot is any good. The next phase will determine whether the UK can turn that competitive dynamic into a coherent framework, or whether the speed of change simply outpaces the capacity to govern it.