Representatives from OpenAI and Anthropic appeared before the UK Business, Innovation, Science and Trade Committee on 13 Oct 2026. Both said the industry’s current self‑regulation does not guarantee safety. Anthropic’s European head, Liam Booth‑Smith, called for “binding rules” on AI companies. OpenAI’s vice‑president, Tom Duff Gordon, agreed that mandatory elements such as incident reporting should be part of a legal framework.
Why the call matters now
The hearing highlighted a split among the biggest AI labs. Google DeepMind and Meta said they would stick to standards rather than new laws, while the US‑based firms pushed for legislation. At the same time, MI5’s director warned that AI could make bioweapon threats more serious, underscoring security concerns.
What the UK currently does
Britain’s AI Security Institute (AISI) can test frontier models only if companies hand them over, and it has no power to stop releases it deems unsafe. The institute’s director, Henry de Zoete, admitted the system is entirely voluntary and dependent on cooperation from US‑ and Chinese‑based firms.
Why it matters
If the UK adopts binding AI safety rules, companies may have to report accidents and submit models for independent testing, which could reduce the chance of dangerous releases reaching the public. For everyday users, clearer regulation could mean fewer unexpected outages or harmful outputs from popular AI services.
