FIS's launch of an AI banking agent for financial crime investigation, reported in Forbes this month, signals something important: AI is no longer experimental in regulated financial services—it is now operational. NICE Actimize's parallel move into agentic capabilities for AML workflows confirms the pattern. For mid-market UK financial services firms, insurers and their legal advisors, this matters because the FCA Consumer Duty (PS22/9) and PRA SS1/23 now require you to justify not just what decisions you make, but how you make them. An AI agent that investigates potential money laundering faster is only valuable if you can explain to the FCA, to your board, and to your clients exactly how it reached its conclusions. Speed without transparency is liability.

This story is part of a broader industry shift: the move from AI-as-analyst (tools that help humans decide) to AI-as-agent (systems that make decisions and take actions with human oversight that may be intermittent or retrospective). FIS and NICE are not alone. Every major fintech infrastructure vendor is racing to deploy autonomous systems into workflows that were previously human-led. The implicit bet is that AI works reliably enough in compliance contexts that humans can move from operator to auditor. But that bet has not yet been tested at scale under real regulatory scrutiny. When it is, the firms that fail will be those that deployed agents first and asked 'how do we prove this was fair and accurate?' later.

Here is Trovix's view: an AI agent is only compliant if it is auditable. That means you need three things. First, you need continuous visibility into how the agent is reasoning—not just what it outputs, but its working. Second, you need a governance framework that sits alongside the agent, not bolted on afterwards. Third, you need documentation that stands up to FCA challenge, Lloyd's Blueprint Two audit, or ISO 42001 assessment. Comparing approaches: Harvey and Legora have focused on legal knowledge grounding; Luminance built its reputation on document pattern recognition. But none of those products solve the governance problem for financial crime investigation, where the stakes for false positives and false negatives are highest. Trovix Audit is designed for exactly this: making AI decision-making transparent and defensible to regulators and internal boards. FIS's agent may be faster, but faster is worthless if you cannot evidence that it complies with FCA expectations on model governance and fairness.

What should you do now? If you are a mid-market financial services firm or an accountancy practice with AML responsibilities, do not wait for your vendor to solve this. Audit your current AI use (even if it is just Copilot or basic chatbots) against the three criteria above: visibility, governance, documentation. If you are deploying or considering an AI agent from FIS, NICE, or any vendor, make governance and auditability a contract requirement, not a post-implementation concern. If you do not yet have a framework for auditing AI decisions at scale, build one now. The FCA will expect it within 18 months, and the firms that get there first will have a competitive advantage in the form of trust.

Source: Forbes