At a British parliamentary hearing, representatives from OpenAI and Anthropic said voluntary pledges are no longer enough to ensure AI safety and called for new laws. Anthropic's Liam Booth-Smith told MPs that rules should be binding, while OpenAI's Tom Duff Gordon supported a statutory framework but emphasized mandatory incident reporting as a starting point. Their position put them at odds with Google DeepMind and Meta, which declined to back new legislation, exposing a deep divide among frontier AI firms.
The hearing also revealed the limits of the UK's current regulatory regime. Civil servants admitted that the AI Safety Institute relies entirely on voluntary cooperation from companies and cannot prevent a model from being released if it is judged too dangerous. AISI director Henry de Zoete said the institute is not in the business of certifying models, and noted that the firms building at the frontier are American or Chinese, limiting what the UK can do alone.
Committee chair Liam Byrne said the divide proves that voluntary self-regulation cannot be trusted. He plans to use the UK's upcoming G20 presidency to push for an international convention that does not depend on the current White House. The session also heard from former lab researchers who said companies cannot be trusted to self-regulate, and separately MI5's director warned that AI could change the stakes on bioweapons. Byrne said the risks are multiplying faster than the safety regime can control.