AI powerhouses OpenAI and Anthropic are calling for binding government regulations in the UK, directly challenging tech giants like Google DeepMind and Meta, which are advocating for self-regulation. The stark contrast emerged during a British parliamentary hearing where AI safety was under the microscope.
During a session with the Business, Innovation, Science, and Trade Committee, Liam Booth-Smith, Anthropic's head of Europe, stated, "The rules should be binding," and that voluntary commitments are insufficient for AI safety. Tom Duff Gordon, VP at OpenAI, echoed a similar sentiment, backing a statutory framework and international convention, particularly emphasizing mandatory incident reporting.
However, their rivals presented a different approach. When asked about supporting new legislation, Google DeepMind's Owen Larter focused on "standards," while Meta's Rob Sherman pointed to existing regulatory bodies in Britain, suggesting they weren't starting from scratch.
This divergence highlights a growing divide in how leading AI firms view their responsibilities. Select committee chair Liam Byrne aimed to expose these differing opinions, stating, "you couldn’t rely on voluntary self-regulation." The hearing also occurred as Britain's security service director, Ken McCallum, warned of AI's potential misuse in creating bioweapons, adding urgency to the regulatory debate.
The session revealed that the UK's current AI safety regime is largely voluntary. Henry de Zoete of the AI Security Institute (AISI) admitted that the institute relies on companies willingly sharing their advanced AI models for testing, and has no power to prevent the release of potentially dangerous technology. Cat Little, a senior civil servant at the Cabinet Office, confirmed the voluntary nature of the current system.
Compounding the issue, executives from OpenAI and Anthropic indicated that access to their models for UK regulators is often facilitated through their US counterparts, suggesting British influence is secondary. Former AI lab researchers also voiced concerns, with ex-OpenAI staffer Daniel Kokotajlo stating, "I don’t think the companies can be trusted to self-regulate," and Steven Adler expressing a lack of confidence in the industry's safety culture.
