OpenAI and Anthropic support binding rules at UK parliamentary hearing
OpenAI and Anthropic backed statutory AI safeguards at a UK parliamentary hearing, while Google DeepMind and Meta argued for different approaches. Former researchers questioned company safety decisions. Officials also described the voluntary arrangements governing access to unreleased models, placing the limits of Britain’s current testing system alongside the companies’ competing regulatory positions.
Artificial Intelligence··Midday
Developers disagree over binding safety rules
Britain’s Business, Innovation, Science and Trade Committee held an AI evidence session on October 13, hearing company representatives and former researchers. The testimony covered safeguards for increasingly capable systems and the risks companies accept during development. The parliamentary session put competing positions on oversight before MPs; it was a hearing rather than the enactment of new legal requirements.[1], [2]
Anthropic’s European head Liam Booth-Smith argued that voluntary commitments were insufficient and supported binding requirements. OpenAI vice president Tom Duff Gordon also backed a statutory framework and an international convention, naming incident reporting as a possible starting point for mandatory duties. Google DeepMind representative Owen Larter stressed standards. Meta’s Rob Sherman pointed to the regulators Britain already has, offering a different route from the new binding framework advocated by the other firms.[1]
Model testing depends on voluntary access
Officials described the operating limits of Britain’s AI Security Institute, the body that examines advanced models before their release. Director Henry de Zoete said access depends on developers and foreign governments. Cabinet Office permanent secretary Cat Little confirmed the voluntary basis of those arrangements. De Zoete explained that the institute neither certifies a model’s safety nor decides whether developers may release it. Testing access and the power to authorize a launch are therefore different responsibilities under the present system.[1]
Duff Gordon said OpenAI would continue supplying predeployment access while the US government permits it. Booth-Smith said Anthropic had previously been unable to provide unrestricted access to Mythos 5.1, but added that the British institute now has access. Committee chair Liam Byrne said the panel would examine Britain’s options and pursue an international convention through the G20 group of major economies.[1]
Former researchers question company safety decisions
Former OpenAI researchers Daniel Kokotajlo and Steven Adler challenged company self-regulation during the hearing. Kokotajlo argued that developers were pursuing systems they did not know how to control. He alleged pressure on staff to soften public comments and restrictions on publishing some internal research. Adler called for governments to set clearer limits on risks imposed by companies. He acknowledged that OpenAI’s internal culture could have changed since his departure in late 2024.[2]
