FCA Warns of Regulatory Overhaul as AI Agents Move Toward Tokenized Money

The UK Financial Conduct Authority (FCA) has published a wide-ranging blueprint for how retail financial services should be regulated as “agentic” AI pushes firms toward near-total automation. In its landmark report, “AI and the future of retail financial services”, FCA executive director Sheldon Mills argues that the industry is moving away from human-led, episodic decisions and toward continuous services delegated to AI systems.

The review, issued as regulators grapple with the speed of generative AI deployment and growing experimentation with blockchain-based finance, frames settlement infrastructure as a key constraint. It suggests that advanced automation will require financial plumbing capable of processing transactions instantly and reliably—something the FCA implicitly contrasts with slower legacy settlement processes.

Key takeaways

• The FCA says retail financial services are shifting from human-led decisions to AI-enabled, continuous and delegated activity.
• The report calls for regulatory work that supports “agentic finance,” including trusted agent protocols.
• It highlights governance and accountability challenges that emerge when AI systems operate autonomously on consumers’ behalf.
• The FCA’s research indicates consumer openness to AI-assisted choices is already material, with 20% of UK adults reportedly willing to let AI make autonomous financial decisions.

An “autonomy spectrum” for retail finance

At the center of the Mills Review is what the FCA describes as an “autonomy spectrum,” reflecting how AI capabilities are evolving from recommending actions to executing them. Mills argues that as models become more independent, humans may increasingly shift into a more passive role—sometimes acting only as observers while AI continuously manages capital.

In the report’s framing, this evolution is not merely about better predictions. Instead, it is about delegation: systems that can be empowered, trained, and authorized to act. That creates a regulatory problem that is different in kind from earlier AI use cases, because the risk no longer rests only with advice or decision support. It rests with operational authority.

The FCA also links the acceleration of this shift to the pace of generative AI. The review notes that more than 20 “frontier models” have been released since late 2025, suggesting firms are moving faster than prior regulatory timetables.

Why settlement speed becomes a regulatory issue

The report connects autonomy to execution. The FCA says that for AI agents to carry out multi-layer transaction strategies smoothly, they require programmable and near-instant settlement mechanisms. By contrast, traditional processes with multi-day latency are described as an operational bottleneck for fully automated finance.

In the FCA’s discussion, systemic stablecoins and tokenized assets are positioned as a potential fit for this need. Because these instruments can be natively issued and transferred on programmable ledger networks, they may enable atomic settlement—where transactions are coordinated in a way that reduces friction and dependency on human clearance.

Importantly, the review does not claim that tokenized settlement is automatically “the answer.” Rather, it highlights a structural challenge: automation at the agent level will stress-test existing financial infrastructure that was built for human workflows and periodic actions.

Accountability, governance, and the “human on the hook” principle

Automation at retail scale also raises questions about who is legally responsible when agents act in unexpected ways. The Mills Review warns that allowing autonomous systems to make and execute decisions introduces severe corporate governance risks, particularly around legal accountability.

The FCA notes industry anxiety about the ambiguity of intention—whether it is possible to reliably distinguish human intent from algorithmic behavior once systems can act continuously and at speed. The report references this concern through industry commentary, including the view that the sector may eventually need something akin to a “Turing test” to separate human intent from machine-driven actions.

In separate remarks, Mills told the Financial Times ahead of the report’s release that accountability must remain anchored to humans. According to his comments, “You need a human on the hook for what they’re doing,” reflecting the FCA’s broader emphasis that operational delegation must not eliminate responsibility.

The Payments Association CEO Emma Banymandhub echoed the governance theme in a statement, saying the FCA’s review “reinforces that firms should treat agentic AI as an accountability and governance issue now,” while maintaining that governance, clear accountability, and consumer trust will determine whether AI’s potential can be realized responsibly.

Recommendations for “agentic finance” and the FCA’s AI capabilities

In its 147-page review, the FCA sets out seven recommendations it says should inform how it responds to the next phase of AI in retail financial services. Among them is an explicit push toward enabling “the foundations for agentic finance”—work intended to support trusted agent protocols that underpin how agentic AI can be used safely.

The report also flags the FCA’s internal capacity, recommending that it consider scaling up its AI Lab to support AI models and system innovation in financial services. The underlying message is that regulatory capability has to evolve alongside the technology it is designed to oversee.

That aligns with the FCA’s earlier steps. The regulator launched the review in January to examine the implications of advanced AI for consumers, retail financial markets, and regulatory oversight (as noted in the FCA’s January press materials).

Still, the report’s most practical impact may be indirect: by framing agentic AI as a continuous, delegated operating model, it implicitly increases pressure on firms to rethink not just their AI tools, but their end-to-end control framework—from authorization and monitoring to dispute handling and accountability trails.

What to watch next

As the FCA turns its recommendations into concrete regulatory expectations, the key uncertainty is how it will balance innovation with enforcement—especially in cases where AI agents execute transactions at speed. Investors and builders should watch for clearer standards on governance, accountability, and how (or whether) tokenized settlement infrastructure fits into the FCA’s definition of safe, trusted automation.