Tom Duff Gordon, OpenAI's EMEA policy head, calls on the UK to introduce binding AI rules to "establish durable, mandatory, capability-based requirements"

If you develop or deploy advanced AI models, the U.K. government may soon require mandatory safety testing and cybersecurity measures.

OpenAI's head of policy for EMEA, Tom Duff Gordon, has urged the U.K. to implement binding regulations for advanced AI development. He stated that the U.K. government should leverage the current political climate to establish "durable, mandatory, capability-based requirements" for frontier AI models, building upon existing voluntary commitments. Duff Gordon emphasized that such legislation would align the U.K. with the EU and emerging U.S. frameworks and position it as a leader in AI safety. The proposed rules should focus on national security, cybersecurity, independent third-party testing, and incident monitoring for a limited number of high-capability AI labs and models. He expressed confidence in the U.K.'s AI capabilities and its potential to influence global AI governance.

OpenAI's public support for binding regulation signals a strategic move to shape the future AI governance landscape, potentially influencing other nations to adopt similar stringent requirements. By advocating for capability-based rules, the company positions itself as a proactive leader in responsible AI development, contrasting with past criticisms of its rapid advancement. This stance suggests a recognition that voluntary measures are insufficient to address escalating safety concerns, particularly in light of recent high-profile incidents and expert warnings.

The U.K.'s adoption of such legislation could create a competitive advantage, attracting AI research and development that adheres to robust safety standards. It also presents an opportunity for the U.K.'s AI Safety Institute to become a central player in global AI oversight, driving international interoperability. The company's emphasis on a narrow focus on frontier models indicates a desire to avoid overregulation that could stifle innovation in less critical AI applications, while ensuring that the most potent AI capabilities are subject to strict scrutiny.