FIS has launched an AI banking agent built with Anthropic to investigate financial crime, while new research shows that 84% of UK and US compliance leaders are concerned about AI-enabled financial crime itself. This is not a contradiction — it is the state of play. Mid-market UK banks, insurers and their legal advisers face an impossible-sounding mandate: cut the $61 billion annual compliance bill while simultaneously managing the new risks that AI creates. The FCA's Consumer Duty (PS22/9) already demands that firms act to prevent harm. Add AI agents that no one fully trusts into that equation, and compliance teams face a governance problem they have no precedent for solving.

What this story reveals is that the industry has reversed its thinking. Two years ago, the conversation was 'AI will save compliance.' Now it is 'AI might break compliance.' The shift is rational. AI agents sound autonomous and objective. They are not. They hallucinate. They invent evidence trails. They replicate the biases in their training data — which in financial crime data means they can amplify historical discrimination. Worse, when they fail, the liability does not disappear; it redistributes. Your firm is still responsible under PRA SS1/23 and FCA rules, even if the agent made the call. FIS's move is smart marketing. But it is not a solution to the problem Ripjar's research uncovered: firms are deploying AI faster than they can audit it.

Trovix's view is this: AI agents should not lead your compliance function. They should augment it, which means you need governance in place first. That sounds slow. It is not. It is the only speed that survives regulatory scrutiny. Trovix Audit exists precisely because firms deploying agents like FIS's need to know what their AI actually does in real time — not what the vendor promised it would do. You need to log decisions, track the evidence chain, flag hallucinations, and prove to the FCA that your AI is not a black box. This is different from how most vendors position AI: Harvey, Legora and Luminance all sell the dream of autonomous legal and compliance work. They are useful tools. But they are tools, not trustees. The distinction matters under UK GDPR and ISA UK. You cannot delegate accountability to software; you can only document how you supervise it. Trovix's approach starts with that supervision layer, not with the agent.

For a mid-market law firm, bank, insurance broker or accountancy practice, the practical move is this: before you deploy any AI agent into compliance or financial crime work, audit your current manual process. Map the decision points. Identify where you cannot afford a mistake. Then introduce AI into only those tasks where you can log, review and reverse its work in real time. Do not expect the agent to learn or improve on its own; it will not, and claiming it has will fail any regulatory interview. Use Trovix Watch to track FCA and PRA guidance as the regulatory framework around AI agents hardens — it will harden this year. And treat the vendor's assurance that the model is safe with the skepticism it deserves. FIS's agent may be better than most. But 84% of compliance leaders are not wrong to worry.

Source: Forbes