Tom Duff Gordon, OpenAI's head of policy in EMEA, calls on the UK to introduce binding AI rules to “establish durable, mandatory, capability-based requirements”
Joseph Bambridge /Politico:
Context & Ripple Effects
The UK had told technology companies in 2024 that an AI bill would turn voluntary commitments into law and put the AI Safety Institute on a statutory footing. Duff Gordon’s intervention pushes that unresolved framework toward capability-based legal duties rather than voluntary commitments alone.
OpenAI’s stance is notable against Sam Altman’s 2023 warning that the company might stop operating where it could not comply with the EU AI Act. It also converges with Dario Amodei’s June call for mandatory third-party testing of frontier-model risks, while Altman has separately advocated a US-led forum for AI standards.
First-order effects
- OpenAI publicly commits its EMEA policy operation to binding UK rules keyed to model capabilities, giving UK lawmakers a clear industry-backed case for legislation.
- Politico’s Joseph Bambridge argues that OpenAI’s support gives Labour an opening to revisit its 2024 manifesto pledge to legislate on AI.
Second-order effects
- A capability-based UK regime would require frontier-model developers to show which systems trigger stronger obligations, moving compliance debates beyond a common voluntary baseline.
- OpenAI’s position narrows the policy distance between it and Anthropic on mandatory safety oversight, increasing pressure on other frontier developers to state whether they support comparable requirements.
Third-order effects
- If UK legislation follows the capability-based approach, safety obligations would increasingly become a condition of deploying the most capable models rather than an optional industry commitment.
- The story points toward AI labs seeking to shape enforceable national and international rules, including standards that can travel across allied markets.
The trend: Frontier AI governance is shifting from voluntary pledges toward capability-triggered, externally enforceable safety requirements.