Anthropic released ten AI agents last week designed to handle financial services tasks: drafting pitch decks, reviewing financial statements, and escalating compliance cases. For UK mid-market firms—law practices, insurers, asset managers, accounting partnerships—the headline sounds like relief. But here is what matters: escalating a case is not the same as reviewing it for compliance. Under FCA Consumer Duty PS22/9, PRA SS1/23, and the SRA Code of Conduct for Solicitors, responsibility for compliance decisions remains with the firm, not the AI system. Anthropic's agents can prepare materials and flag issues. They cannot discharge your regulatory duty. The distinction is not semantic.
This story is part of a wider pattern we see repeating across the industry. Vendors—Anthropic, Harvey, Legora, Microsoft—are racing to deploy agentic systems that appear to remove human judgment from regulated workflows. The market pressure is real: faster document throughput, lower cost per review, 24-hour availability. But the regulatory architecture has not changed. The ICO's AI guidance, the EU AI Act's Article 6 high-risk classification, and the emerging ISO 42001 standard all point in the same direction: human oversight is not optional in high-stakes decisions. AI agents that look autonomous are often just postponing accountability.
Trovix's view is this: the firms winning with AI are those that use it to enhance human decision-making, not replace it. Our approach to compliance automation—visible in Trovix Audit—is built on the principle that AI surfaces risk and context, but qualified humans make the call. Anthropic's agents can certainly draft a pitch deck or summarize a financial statement better than spreadsheets. But if you deploy them to 'escalate compliance cases', you have created a system where compliance escalation is delegated to a system that cannot be held accountable. That is not compliance automation. That is compliance theatre. The difference between a tool and a proxy is whether you can explain every decision it makes to the FCA.
What should a mid-market financial services firm, law practice or insurer actually do? First, audit what you are currently doing manually in compliance workflows. Trovix Watch helps track where regulatory requirements are evolving. Second, deploy AI agents only where they genuinely reduce friction without removing accountability—like intake triage, document classification, or preliminary summarization. Third, keep a human-in-the-loop protocol wherever the AI outcome touches a regulatory decision. Fourth, document the AI's role in your governance framework; your compliance officer and external auditor need to see the decision tree. Anthropic's agents are technically impressive. But regulatory firms do not compete on technical impressiveness. They compete on defensibility.
Source: Bloomberg News