OpenAI and Anthropic Back Binding AI Safety Rules at UK Hearing

OpenAI and Anthropic told a British parliamentary committee that voluntary AI safety pledges were no longer sufficient and called for statutory regulation. At the hearing, Anthropic's head of Europe Liam Booth-Smith said rules on AI safety should be binding and that voluntary commitments are not enough, according to Wired.

OpenAI vice president Tom Duff Gordon backed a statutory framework and an international convention, saying that some elements, starting with incident reporting, should be mandatory. Not all witnesses took the same position. Google DeepMind's Owen Larter did not back a statutory framework, saying the focus should be on standards, while Meta's Rob Sherman pointed to existing British regulators and told MPs they were not writing on a blank page.

Limits of the current voluntary regime

Cabinet Office permanent secretary Cat Little told the committee that the UK's AI regulatory regime is voluntary at the moment. AI Security Institute director Henry de Zoete acknowledged that the institute relies on companies voluntarily handing over frontier models and cannot prevent the release of a model it deems too dangerous.

The hearing followed warnings about misuse. MI5 director Ken McCallum used his annual threat update to flag bioweapon misuse as a significant threat from AI, the misuse case he said he would attach the greatest weight to. In previous remarks, Booth-Smith said Anthropic was not able to provide full unrestricted access to a model referred to as Mythos 5, while Duff Gordon said OpenAI provides its models pre-deployment to the US counterpart of the AI Security Institute and will continue to provide them pre-deployment to the UK institute as long as it can under the US government.