Anthropic's decision to IPO despite losing $42 billion in 2025 tells you something important: the venture capital model for AI vendors is exhausted, and public markets are expected to absorb the next phase of losses. For UK regulated firms in law, insurance, financial services and accountancy, this is not a sideshow. When your AI vendor goes public, vendor risk management gets harder, not easier. You will inherit new reporting obligations, shareholder pressures, and potential pivots away from the customer segments you depend on. The fact that Anthropic achieved profitability for a second consecutive quarter matters less than the journey to get there — $42 billion in annual losses tells you the unit economics of renting out Claude models, no matter how good they are, do not yet work at scale. That should concern any firm making long-term procurement decisions.

This story is part of a broader pattern: generalist AI vendors (Claude, GPT, Gemini) are moving toward public markets because they cannot fund themselves through venture capital without hitting the ceiling on what private investors will tolerate. Meanwhile, regulated firms have spent the last 18 months discovering that plugging a generalist AI model into your workflow and hoping for compliance is not a strategy. The FCA's consumer duty (PS22/9), the SRA's code on AI use, the PRA's AI governance expectations (SS1/23), and the incoming EU AI Act all demand that firms understand and take responsibility for how AI makes decisions, especially where client money, legal advice or insurance underwriting is involved. Harvey, Legora and Luminance built from the ground up to be domain-specific because they understood this. Anthropic built a better general-purpose model. Neither approach solves the governance problem that regulated firms now face.

Here is Trovix's honest view: the real business opportunity is not in renting smarter models. It is in helping regulated firms implement AI in a way that survives regulatory scrutiny and does not blow up when the model hallucinates or behaves in ways you did not anticipate. Anthropic going public means Claude will keep getting better and cheaper. That is good for users. But it does not solve the hard problem: integrating AI into processes where you have fiduciary duties, client confidentiality obligations, or anti-money laundering responsibilities. Most AI implementations we see in mid-market firms treat the model as a black box and bolt governance onto the back end. That works until it does not. The alternative — which we have built our platform around — is to make governance and explainability intrinsic to how the AI works. That means understanding the data the model trains on, the guardrails you need in place before it touches client work, and what happens when it fails. Microsoft Copilot in Microsoft 365 offers scale and integration. Claude offers raw capability. Neither gives you the audit trail, the policy framework, or the accountability layer that FRC ISA UK, PRA SS1/23, and the ICO expect you to have.

If you are a law firm, insurer, financial services business or accountancy practice deciding whether to adopt Claude via Anthropic, or weighing it against other vendors: make your decision on capability, but condition your deployment on governance infrastructure that your compliance team can actually defend to a regulator. Treat vendor IPO announcements as a signal to stress-test your AI governance framework. If Anthropic becomes a public company with quarterly earnings pressures, the incentives change. Their obligation shifts from serving you to serving shareholders. That does not make them a bad partner, but it does mean you need to own the governance layer yourself. Trovix Audit exists precisely because procurement decisions like this one create new governance obligations downstream, and most firms are not ready for them.

Source: CNN