OpenAI and Anthropic Tell British Lawmakers: Regulate Us

Your access to advanced AI models may become more restricted or subject to new reporting requirements.

During a session with British lawmakers, AI companies OpenAI and Anthropic advocated for mandatory government regulation of artificial intelligence, including statutory frameworks and binding rules. OpenAI's vice president suggested mandatory incident reporting as a starting point, stating that voluntary commitments are insufficient for managing frontier risks. In contrast, Google DeepMind and Meta expressed a preference for focusing on standards and leveraging existing regulatory bodies, revealing a division within the industry on AI governance. The hearing also highlighted concerns from Britain's security service about AI's potential misuse in bioweapons. Civil servants admitted that the UK's current AI safety testing regime is voluntary, relying on company cooperation, and that they lack the power to prevent the release of dangerous models. Executives indicated that their cooperation with UK bodies is often contingent on US government approvals, underscoring the UK's limited autonomy in regulating these frontier AI models.

The stark contrast between OpenAI/Anthropic's call for binding regulations and Google/Meta's preference for voluntary standards signals a widening chasm in the AI industry's approach to governance. This divergence, publicly aired before a parliamentary committee, suggests that future regulatory battles will be fought not just between industry and government, but also between competing factions within the AI sector itself. The UK, acting as a middle power, now faces the challenge of charting an independent regulatory course, potentially through international alliances, to address AI risks that transcend national borders and current geopolitical influences.

The admission that the UK's AI safety regime is entirely voluntary and dependent on foreign entities, particularly the US, exposes a critical vulnerability in the nation's ability to safeguard against advanced AI threats. This reliance raises questions about data sovereignty and the practical implications for British companies and researchers seeking access to cutting-edge models. The revelation that regulatory power is not absolute and hinges on cooperation from major AI labs and their home governments suggests that the UK's influence on global AI safety standards may be more limited than previously assumed.