Anthropic released ten AI agents last month designed to handle financial services work — pitch decks, financial statement review, compliance escalation. The market reaction was lukewarm, which tells you something important: the City knows that raw capability is not the same as regulatory readiness. For a UK financial services firm, insurance broker or accountancy practice, the question is not whether Anthropic's tools can do the work. The question is whether you can defend the decisions they make to the FCA, PRA, SRA or ICO. Anthropic's announcement sidesteps that entirely. The agents exist. The governance framework does not.

We are watching a predictable pattern repeat. First, a major AI lab (OpenAI, Google, Anthropic) releases a capability that sounds transformative. Second, it gets positioned as plug-and-play for regulated industries. Third, regulated firms discover that 'the AI did it' is not an acceptable explanation to regulators, and that the vendor's training data, decision logic and audit trail are either opaque or incomplete. Harvey, Legora and Luminance have all built products specifically for legal work because they understood that lawyers operate under professional liability and regulatory oversight. But in financial services, the gap between capability and defensibility is wider still. You are handling client assets, regulatory reporting, market conduct obligations and anti-money-laundering controls. The FCA's Consumer Duty PS22/9 and the approach to algorithmic accountability in the EU AI Act both demand that you can explain what happened and why. A black-box agent that drafts a pitch deck or escalates a compliance case is not enough.

Here is Trovix's honest take: general-purpose AI agents from capability labs are not fit for regulated financial services work without significant additional layer. You need governance scaffolding — audit trails, decision checkpoints, output validation, and clear handoff points back to qualified humans. You need to know what data the agent was trained on, what it can and cannot do reliably, and what human review processes close the accountability gap. Trovix Audit exists precisely because we saw firms deploying AI without that framework and getting into regulatory trouble. The agents themselves can be valuable. The governance envelope around them is what matters. Anthropic's announcement is honest about capability and silent about governance — which is exactly backwards.

If you are a partner at a mid-market law firm, an underwriting manager at a broker, or a finance director at an accountancy practice, do not assume that Anthropic's agents (or similar general-purpose tools) are ready for your regulated work. Before you evaluate any AI agent for compliance, document review or financial analysis, ask: Can we audit every decision it makes? Can we explain it to a regulator? Do we have a human review process that catches errors? What happens when it gets it wrong? If the vendor cannot answer those questions with specificity, the tool is not ready. Build the governance framework first. Then, and only then, evaluate whether the capability is worth the integration cost.

Source: Bloomberg News