Representatives from OpenAI and Anthropic addressed a British parliamentary hearing, stating that voluntary pledges are insufficient for ensuring safety in artificial intelligence (AI) and advocating for new legislation. This hearing, conducted by the Business, Innovation, Science, and Trade Committee, included participation from four AI firms, whistleblowers, and senior civil servants.
Liam Booth-Smith, head of Europe at Anthropic, emphasized the need for binding regulations, asserting that voluntary commitments alone are inadequate. OpenAI's vice president, Tom Duff Gordon, supported the idea of a statutory framework and suggested that certain elements, such as incident reporting, should be mandatory.
In contrast, representatives from Google DeepMind and Meta did not endorse new legislation. Owen Larter from Google DeepMind indicated a focus on standards rather than statutory frameworks, while Rob Sherman from Meta referred to existing regulatory structures.
The hearing highlighted a divide among AI leaders regarding regulation, as noted by committee chair Liam Byrne, who aimed to showcase the differing opinions among major firms.
Simultaneously, MI5 director Ken McCallum warned of the potential misuse of AI in bioweapons during his annual threat update. Byrne remarked on the contradiction between the MI5 warning and the voluntary nature of the current regulatory regime.
Civil servants acknowledged the limitations of the existing system, with the AI Security Institute (AISI) relying on voluntary cooperation from companies for testing frontier models. Henry de Zoete, director of AISI, stated that they do not have the authority to prevent companies from releasing models deemed too dangerous.
Duff Gordon explained that OpenAI provides models to AISI in the UK, contingent on US government regulations. Booth-Smith highlighted issues with providing unrestricted access to certain models, attributing this to US government policies.
The session also featured insights from former lab researchers who expressed concerns about the industry's self-regulation capabilities. Following the hearing, Byrne indicated that the committee would focus on the UK's future role in AI regulation, aiming to leverage Britain's upcoming G20 presidency to foster an international convention on AI safety. He noted that the risks associated with AI are growing faster than the current safety measures can address.