Anthropic's announcement of ten specialist AI agents for financial services—capable of drafting pitch decks, reviewing financial statements and flagging compliance issues—represents real technical progress. For mid-market banks, insurers and asset managers under FCA oversight, the appeal is obvious: faster document processing, reduced manual review, lower operational cost. But the story masks a critical gap. None of these tools address what actually keeps a compliance officer awake at night: proving to the FCA that your AI is trustworthy. Under PS22/9 (Consumer Duty) and the emerging EU AI Act framework now informing UK regulatory thinking, a tool that works well is not enough. You must demonstrate how it works, why it fails, when it hallucinates, and who remains accountable when it does. Anthropic has built the engine. It has not built the control panel.

This story is part of a larger pattern. Harvey, Luminance, and now Anthropic are racing to verticalize AI—to build tools that speak the language of specific industries. This is sensible product strategy. But the industry is converging on a shared blind spot: agentic systems are being positioned as time-savers when the actual constraint for UK regulated firms is governance risk. The FCA's 2025 final rules on AI governance (expected later this year) will require documented risk assessments, ongoing monitoring, and clear lines of accountability. Tools that automate without instrumenting governance compliance make firms less efficient at compliance, not more. The winners in financial services AI will not be the fastest tools. They will be the tools that collapse the time between execution and auditability.

Trovix's view is straightforward: financial services firms should be deeply sceptical of agentic tools that do not come paired with governance visibility. An AI agent that reviews financial statements in seconds but generates no audit trail of its reasoning is a liability dressed as productivity. This is why we built Trovix Audit as a governance dashboard, not an agent showcase. When you deploy Trovix products—whether Aria for knowledge retrieval or Sift for document extraction—every decision is logged, traced and defensible. We have learned from watching competitors (and from seeing firms regret their choices) that regulated firms do not want agents that work in the dark. They want transparency. Anthropic's agents work. Ours work too. The difference is that when the FCA calls, you can show your working.

If you are a mid-market law firm, insurer, asset manager or accountancy practice evaluating financial services AI right now, do this: ask every vendor the same question. When your system makes a decision, can I see exactly why? If the answer includes the word 'proprietary' or 'black box', walk away. You do not have the scale or the in-house AI engineering team to manage that risk. Anthropic's May announcement is technically impressive. But it is an accelerant on a problem—governance complexity—that it does not solve. The practical move is to pilot tools with built-in auditability first. Speed will follow. Compliance will not have to chase it.

Source: Bloomberg