On 5 May, Anthropic released AI agents designed to handle financial services workflows — pitch decks, financial statement review, compliance case escalation. The market response was sharp: FactSet shares fell 8.1%, a sign that the City sees real disruption incoming. For mid-market UK regulated firms, this matters immediately. The FCA's Consumer Duty (PS22/9) makes firms responsible for the outcomes of tools they deploy. The PRA's recent SS1/23 on operational resilience requires firms to understand third-party AI risks. Anthropic's agents are built for task execution, not for the audit trails, decision logging, and compliance documentation that regulators actually demand. That gap is not a technical limitation. It is a business model choice.
This story is part of a larger pattern: AI vendors are racing to productize agents for regulated workflows, but they are treating compliance as a feature layer bolted onto capability layers. Harvey built legal AI and then added compliance workflows. Luminance and Legora focused on document intelligence first, governance second. Microsoft Copilot for financial services follows the same pattern — raw capability, then scaffolding for control. The pattern works in unregulated contexts. It fails catastrophically in regulated ones. The FCA does not care what the agent can do. It cares what you can prove it did, why it did it, and who signed off on it. When a pitch deck drafted by an Anthropic agent contains a material omission, or a financial statement review misses a red flag, the firm deploying it is liable — not the vendor. Anthropic's agents have no mechanism to log that liability.
Trovix's view is this: agentic AI in regulated firms must be built backwards from governance, not forwards from capability. That means starting with audit-ready logging, decision transparency, and human sign-off workflows — then wrapping capability around them. It means treating the agent not as a replacement for professional judgment but as a documented step within a controlled process. Trovix Audit was built on this principle: every agent action is timestamped, attributable, and reviewable. Every escalation is logged. Every override is recorded. This is not faster than Anthropic's agents. It is compliant. The broader market will learn this the hard way — through FCA enforcement actions against firms that deployed Anthropic's agents without adequate governance infrastructure. Trovix's bet is that smart firms will learn it now.
What should a mid-market law firm, insurer, or financial services business do? First, do not assume that a product working well in beta testing will work at scale under FCA scrutiny. Second, if you are evaluating Anthropic's agents or similar products, ask the vendor for their audit and escalation logs from three months of production use. If they cannot produce them, walk away. Third, build your own governance layer before deploying any agentic AI — this is not optional under Consumer Duty and SS1/23. If you lack the in-house capability, that is what partners like Trovix exist to provide. The competitive advantage in regulated sectors is not being first to deploy agents. It is being able to deploy them safely and prove it to regulators. Anthropic's launch is a capability milestone. It is not a readiness milestone for regulated use.
Source: Bloomberg News