On 5 May, Anthropic released ten agents designed to draft pitch decks, review financial statements and flag compliance issues for human review. On the surface, this is relevant to mid-market UK financial services and insurance firms facing real capacity constraints in back-office work. The problem is not whether these agents can produce output that looks professional. They can. The problem is that deploying them without compliance governance is a regulatory risk that Anthropic has not solved. Under FCA Consumer Duty PS22/9 and PRA SS1/23, firms must be able to demonstrate that autonomous decision-making — or near-autonomous assistance — is auditable, traceable and subject to human accountability. Anthropic's agents, as described, handle document workflow. They do not handle the audit layer.

This is part of a wider pattern. In 2024 and 2025, every major AI vendor moved toward agent architecture — OpenAI's swarm experiments, Microsoft's Copilot Studio, Legora's workflow automation, Harvey's escalation logic. All of them can execute tasks faster than human teams. None of them started by asking: what does a regulated firm's compliance officer need to sign off on this? The result is a two-tier market forming. Vendors who build capability first and compliance second. And vendors — still relatively rare — who understand that in financial services, insurance and legal, the audit trail is not an afterthought. It is the product.

Here is Trovix's honest view: Anthropic's agents are good at what they do, but they are designed for speed, not governance. If your firm deploys them to draft pitch decks or summarise statements, you will need to layer governance on top — logging which agent version was used, what prompt was input, what human reviewed the output, and crucially, who is liable if it goes wrong. You can do this with Trovix Audit, which sits above any agent architecture and creates the compliance dashboard regulators expect. But you should not have to buy a separate product to make an agent deployable. Harvey and Legora understood this earlier. Anthropic is now playing catch-up on the compliance side, not the capability side.

For a mid-market law firm, insurer or financial services practice, the immediate action is not to wait for Anthropic to fix this or to assume their standard deployment works in a regulated context. It does not. If you want to use AI agents for compliance-sensitive tasks — financial statement review, pitch preparation, case escalation — you need three things: the agent itself, the governance layer that tracks what happened, and a framework for human accountability that your compliance team can audit. Start with your FCA or SRA obligations first. Build the compliance governance second. Then choose your agent vendor. That sequence matters more than which vendor you pick.

Source: Bloomberg News