Anthropic's announcement in May that its AI agents can draft pitch decks, review financial statements and escalate compliance cases is attracting justified attention from UK financial services firms. The toolset targets real workflow problems in banking, insurance, asset management and fintech — areas where regulated professionals are drowning in routine document work. But here is what matters: the FCA's AI rulebook (part of which sits within PS22/9 on Consumer Duty, with more coming via PRA SS1/23 on operational resilience) does not care whether your AI makes decisions autonomously or with human sign-off. It cares about who is accountable when something goes wrong. Anthropic has released agents. It has not released a liability framework. UK mid-market firms need to understand the difference before deployment.

This story is symptomatic of a wider pattern in the industry: vendors are racing to build agentic systems — autonomous tools that take actions without real-time human intervention — while the regulatory framework for that autonomy remains fragmented. Harvey, Luminance and others have spent years building document-intelligence products that augment human decision-making. Those products sit comfortably within established compliance models because a lawyer or compliance officer remains the decision-maker. Anthropic's agents are different. They escalate compliance cases. That escalation IS a decision. Under FCA, PRA and SRA frameworks, someone must be responsible for that decision's quality and correctness. The question is: who? And what happens when the agent gets it wrong?

Here is Trovix's honest view: AI agents will absolutely be part of the future of regulated professional services. But they cannot be deployed as black boxes. The firms that will succeed are those that treat agents as one component within a documented, audited governance system — not as a replacement for governance. This is why Trovix Audit exists. You need end-to-end visibility of how AI is making or influencing decisions, audit trails that survive regulatory review, and a clear ownership chain. Anthropic's tools are technically sophisticated. But 'we deployed Anthropic agents' is not a defence in an FCA investigation. 'We deployed Anthropic agents within an ISO 42001-aligned governance framework with documented escalation rules, human override capability and monthly audit reviews' is.

What should your firm do now? First, do not treat this announcement as a signal to rush into deployment. Second, audit your current AI tools to understand where real autonomy already exists — many firms using document-intelligence platforms do not realise they are already making quasi-autonomous decisions at scale. Third, if you are serious about agents, demand from any vendor (Anthropic or otherwise) a clear accountability model and audit integration before you sign. Fourth, brief your compliance and risk teams before your technology team. The regulatory burden of agent deployment falls on compliance. Make sure they understand what they are inheriting.

Source: Bloomberg News