OpenAI Backs a UK AI Law Aimed at the Most Powerful Labs
OpenAI urged British lawmakers on Sept. 14, 2026, to pass new laws to control AI technology, just after Anthropic called for limits on AI development. The Guardian
The request came from Tom Duff Gordon, OpenAI's head of policy in Europe. OpenAI said: "We support stronger UK rules for the handful of companies, including OpenAI, developing the most powerful AI systems."
OpenAI said the law should cover only the most advanced AI labs, not startups working on less powerful systems. Like an emissions rule aimed at the biggest factories, it would place legal duties on the few teams training the largest models and leave smaller builders outside.
Duff Gordon told Politico the UK government should "take advantage of this political window which is clearly opening up" to pass AI legislation. He was pointing to debate in Westminster on the same day.
That debate centers on the cross-party Joint Committee on Human Rights, a group of MPs and Lords. Led by Labour MP Alex Sobel, it called for a new AI framework with an independent oversight body and laws to protect the public. It said a new AI bill should set tougher duties for higher-risk AI systems, meaning uses where harm to people or rights is more likely.
The committee published its report, 'Human Rights and the Regulation of AI', on Sept. 14, 2026. It followed an inquiry, 'Human Rights and the Regulation of Artificial Intelligence', launched in July 2025 to look at what rules are needed to protect human rights when AI is built and used. UK Parliament
In the weekend before Sept. 14, Anthropic had urged rivals and governments to coordinate a global slowdown in AI development. OpenAI, Google DeepMind and Elon Musk supported that call. Separately, Anthropic researchers have warned that artificial intelligence could kill all of humanity by 2030.
More than 70 MPs and peers urged Prime Minister Andy Burnham to support a ban on creating artificial superintelligence (ASI), a hypothetical future AI far more capable than people. Downing Street rejected an ASI ban but said it was exploring additional targeted interventions "to address the most significant AI-related national security risks".
A parallel debate is underway in Washington. OpenAI urged the U.S. Congress to adopt capability-based national AI safety requirements, meaning rules tied to what a system can do, according to reporting published Sept. 10, 2026. Those proposals include testing standards and independent assessments. Reuters
Britain's data watchdog said on Aug. 3, 2026, that it was watching developments closely involving OpenAI and Anthropic after rogue AI agent hacking incidents. An agent here means AI software that can take actions on its own. That monitoring is separate from the human-rights proposals before Parliament, but shows regulators are also watching how deployed tools behave.
Earlier history matters. OpenAI CEO Sam Altman told the U.S. Senate Judiciary Committee he supports guardrails, or basic safety limits, to reduce harms from AI. OpenAI told a UK committee it would be impossible to build tools like ChatGPT without access to copyrighted material. Anthropic spent $910,000 on lobbying in the second quarter, up from $150,000 in the same period last year, as reported by Politico.
The broader context here is a move from voluntary promises to written law. London has relied on existing sector regulators and principles-based guidance. The Joint Committee plan, together with industry calls for binding rules on the largest labs, points to a tiered system where duties rise with capability and risk.
Looking at what this means for diplomats and policy planners, three questions will shape what happens next. First, how Parliament defines the handful of companies and higher-risk systems facing tougher duties. Second, whether a UK oversight body could act before a system launches or only enforce rules afterward. Third, how any UK bill fits with a U.S. capability-based standard and with Anthropic's proposed global slowdown, which would need coordination beyond one country.
