Senator Elizabeth Warren has it right. AI companies asking Congress for antitrust exemptions to collaborate on safety are asking the wrong question — and mid-market UK regulated firms should pay attention. The story CNBC reported yesterday shows major AI vendors lobbying for carve-outs to work together on 'safety standards', claiming competition law blocks the collaboration they need. Warren's response is sharp: antitrust law already permits safety collaboration. What it does not permit is market-sharing disguised as safety. For UK firms subject to the FCA Consumer Duty, SRA Code, PRA SS1/23, and the emerging EU AI Act equivalents, this distinction matters enormously. Your AI vendor is not exempted from competition law because they call their arrangement 'safety-focused'. Neither are you.

This story is a symptom of a deeper misalignment in the AI industry. Vendors like Harvey, Legora, and Luminance have built their value propositions on proprietary training, closed-loop data handling, and competitive moats dressed as safety features. They now discover that claiming 'safety' as justification for refusing interoperability, refusing to publish model cards, or refusing to collaborate with competitors does not convince regulators — because regulators care about actual safety outcomes, not vendor convenience. The pattern is clear: when an industry cannot solve a problem through transparent competition, it asks for regulatory exemption. Regulators are right to refuse. The shift towards mandatory transparency (EU AI Act Article 13, ICO UK GDPR risk assessments, FRC ISA UK audit evidence standards) means vendor safety claims must be demonstrable, not merely asserted.

Trovix's approach to this tension is deliberate. We do not ask for exemptions or build products that require them. Trovix Aria is built on retrieval-augmented generation with transparent prompting — it works within existing competition and data protection law without friction. Trovix Sift extracts data without vendor lock-in; your firm retains control. Trovix Watch monitors regulatory change so you stay ahead of this exact problem. We see firms abandoning Harvey deployments because they cannot audit the training data, or hitting walls with Copilot because Microsoft's data handling creates GDPR exposure, or finding Luminance's opacity incompatible with SRA audit obligations. These are not edge cases — they are the cost of products built on the assumption that safety and competition are enemies. They are not.

Here is what you should do Monday morning. First, audit your current AI deployment against three questions: (1) Can your vendor demonstrate their safety claims independently, or do they ask you to trust their word? (2) Does your contract give you audit rights over training data and model behaviour, or have you waived those? (3) If your vendor went out of business tomorrow, could you extract your data and continue operations? If you answered 'no' to any of these, your AI is not compliant with emerging FCA, SRA, and PRA standards — it is compliant with vendor interests. Second, when evaluating new AI tools, reject any vendor claiming antitrust or safety exemptions justify their refusal to be transparent. Third, build your procurement around products that assume competition and transparency are features, not bugs.

Source: CNBC