Financial services’ AI risk shifts from answers to actions

Financial services firms are moving from AI that answers questions to systems that take action. The UK Government’s Financial Services Adoption Plan calls on the sector to scale AI across core processes, including fraud investigations, claims and compliance work.

That shift makes it harder to track responsibility. Model explainability is only part of the challenge: firms also need to see which systems an agent accessed, what it did and where a person reviewed or changed its actions. Financial Conduct Authority (FCA) Chief Executive Nikhil Rathi has described the technology’s role as beginning to “coordinate and transact”.

The source reports that 48% of UK financial services executives say their firms use agentic AI, while more than a quarter report having no or limited controls to ensure compliance with laws and regulations. Agents can move among systems and tools, leaving decisions distributed across a workflow rather than clearly assigned.

In an insurance claim, for example, one agent might gather evidence and route a case for closer scrutiny; another might contact the customer or recommend withholding payment. Firms need enough information to reconstruct how an outcome was reached, including the agent’s inputs, tools, rules and any human review.

The source argues that oversight should reflect the consequences of an action. Routine administrative work may need monitoring, while actions such as restricting a bank account, rejecting a claim or escalating a fraud allegation call for stronger human oversight and the ability to challenge or reverse a decision.

AI agents can reduce repetitive work, the source says, but scaling their use depends on clear ownership, defined permissions and records covering the full workflow, including activity involving third-party services.