EU and UK AI Regulation Converging, Not Diverging
Despite fears of regulatory divergence, the EU and UK are actually converging on AI regulation, both focusing on outcome accountability and auditability rather than technology-specific rules.
Many founders worry that the UK and EU are drifting apart on AI regulation. The UK has refused to write a bespoke AI law, while the EU wrote one that few can keep up with. However, beneath the political surface, the two regimes are converging, not splitting.
The UK has not enacted a specific AI statute, instead regulating outcomes through existing frameworks like the Senior Managers regime and Consumer Duty. The EU's AI Act initially set high-risk obligations but delayed their enforcement to end of 2027 because technical standards were not ready. The FCA's Mills Review, published last week, confirms that an outcomes-based approach is the right foundation, and now is not the time to rush new rules.
What determines a company's exposure to AI regulation is not where it is incorporated, but whether it can reconstruct what its AI did when asked by a regulator or customer. Neural network explainability – opening the model to read its reasoning – is often an illusion. Regulators want an evidence trail: which document, which clause, which rule, what the system flagged, and who reviewed it.
This auditability must be built from the first line of code – observability at every step, not just the output. The Mills Review directly states that governance is likely to become an enabler of capability, not a constraint. Firms that demonstrate auditability and explainability will be able to deploy AI more confidently.
The EU AI Act similarly requires automatic logging, pre-market technical documentation, and meaningful human oversight. Like the UK, it supports decisions that can be reconstructed and defended.
Founders should not wait for regulatory clarity. Build systems where every decision can be reproduced and defended, with a human answerable for it. Do that and you're compliant in London, Frankfurt, and most places in between. The rules that matter already apply. The work is not lobbying for a cleaner rulebook, but building and governing AI systems that can answer the one question every regulator asks: what happened, and who is answerable?
