FIS and Anthropic's AI banking agent is now live, but the timing of that Forbes story carries a warning that most UK regulated firms still haven't internalized: when your AI vendor's decision harms a customer, regulators treat the vendor's mistake as *your* mistake. The FCA's Consumer Duty (PS22/9) makes this crystal clear — outsourcing to technology providers does not transfer accountability. If an AI agent misclassifies a customer, flags the wrong transaction, or makes a compliance error, your firm remains liable. This is not hypothetical. The story makes plain that customers don't distinguish between bank and technology provider, and neither will the FCA, PRA, or ICO when they investigate.

What we're watching is the fintech industry learning a painful lesson about the difference between capability and responsibility. Harvey, Luminance, and other legal-focused AI tools have built impressive products, but the ones that survive regulatory scrutiny will be those with auditable decision paths and clear governance frameworks. The broader pattern is this: firms are deploying AI agents without the infrastructure to explain *why* the agent made a decision — and regulators are beginning to demand exactly that. The EU AI Act's requirements for high-risk systems are already forcing European banks to think harder about this. UK firms cannot assume they have time to catch up later.

Here's Trovix's honest take. The problem with most AI agent implementations isn't the AI itself — it's the governance layer. FIS has a technically sophisticated agent; what matters now is whether their banking partners can audit every decision, trace accountability, and demonstrate compliance to regulators in real time. That requires three things most firms lack: a clear audit trail (Trovix Audit exists because firms kept telling us they couldn't find that), continuous monitoring of regulatory change so you know which rules your AI needs to follow (Trovix Watch), and honest documentation of where the AI can and cannot be trusted. Microsoft Copilot in financial services fails at this because it's a general-purpose tool retrofitted to regulated work. Purpose-built systems with governance built in from the start perform better. But even then, your compliance team needs to own the AI, not merely use it.

If you're a mid-market bank, insurer, law firm, or accountancy practice, the immediate action is this: audit what AI you're already running, then reverse-engineer the accountability trail. Can you show a regulator exactly how the AI made each decision? Can you point to where a human remained responsible? If the answer is no, you have a compliance gap that no amount of vendor reassurance will close. Before you deploy new AI agents — whether from FIS, or anyone else — demand that your vendor provides governance tooling as part of the contract, not as an afterthought. The FCA's principles-based approach means you own the compliance obligation, full stop. Act like it.

Source: Forbes