Senator Elizabeth Warren and her allies have rejected requests from AI companies like Anthropic for narrow antitrust exemptions that would allow competitors to share safety research. On the surface, this is ideological pushback against consolidation. In practice, it means that major AI vendors cannot formally collaborate on safety standards, and UK regulated firms have no way to verify whether the tools they are buying—Harvey, Legora, Microsoft Copilot, or any other—meet consistent safety benchmarks. This matters acutely because the FCA's Consumer Duty PS22/9, the SRA's Competence rules, and the PRA's SS1/23 all require firms to understand and manage third-party AI risk. When vendors are contractually prevented from sharing safety findings, that duty becomes impossible to discharge properly.

This row reveals a fundamental contradiction at the heart of AI governance right now. Regulators in the UK and EU demand that firms implement AI responsibly and transparently. Meanwhile, US antitrust doctrine is making it illegal for the companies building AI to collaborate on the very safety practices that regulators expect. The US has chosen competitive ideology over collective safety. Europe's AI Act mandates conformity assessments and risk classification. The UK has opted for principles-based principles. But none of these frameworks can work if the vendors themselves cannot pool knowledge on how to prevent harm. We are watching regulation and commerce move in opposite directions.

Trovix's approach to this problem is different. Rather than betting on vendor claims about safety—which remain untested when vendors cannot collaborate on standards—we help firms implement their own AI governance layer. Trovix Audit gives you independent visibility into how every AI tool in your firm is performing against your own risk appetite and regulatory obligations, regardless of what the vendor claims. You do not wait for an industry-wide safety consensus that may never come. You measure compliance yourself. Trovix Sift and Trovix Aria are built on the premise that if you cannot trust the vendor ecosystem, you need to audit it in real time. Products like Harvey and Legora are competent tools, but without governance infrastructure of your own, you are exposed to regulatory drift and vendor-side risk you cannot see.

Your firm should assume that major AI vendors will not achieve safety consensus in the next 18 months. Plan accordingly. Conduct a real audit of which AI tools are currently in use (most mid-market firms undercount by 60–70 per cent). For each tool, document your use case, the data you are feeding it, and the regulatory obligations it touches. Then build a lightweight but continuous compliance check—not a one-off audit, but a dashboard that flags drift. If you are already using Trovix Reach for client-facing AI, layer governance on top of it. The regulator will not accept a defence that 'the vendor said it was safe.' You will be expected to prove it.

Source: CNBC