AI Tracker
Ex-Ante Regulation Benchmark updated to include a proposal for a law that would ban the development of artificial superintelligence in the UK
On 8 September 2026, in the UK, the Artificial Superintelligence (ASI) Bill was introduced in the House of Commons, which would ban the development, deployment and operation of ASI systems. Although ASI is generally understood as an AI with better cognitive abilities than even the most intelligent humans, a more comprehensive and consistently agreed-upon definition does not yet exist. Calls from UK politicians for a ban on ASI have been, in part, encouraged by fears over the safety of AI development that have become more prevalent in recent months following the OpenAI/Hugging Face cyber incident and statements from AI researchers on the technology’s capacity to put human life at risk, with one Anthropic researcher saying that there is a greater than 10% chance it "could kill all humans" within the next decade.
These concerns have led policymakers and AI firms to call for the technology to be regulated. For example, Jack Clark (Co-Founder, Anthropic) said that AI firms may need to be required to have a “kill switch” for all of their AI operations. While no jurisdictions have imposed such a requirement on AI firms, a bill for an AI Emergency Button Act (which would require such a kill-switch for all AI systems) was introduced in the US Senate on 16 September 2026, highlighting that although novel, this approach is being considered. Our Ex-Ante Regulation Benchmark in the AI Tracker shows that the closest in-force approach to this has come from the EU’s AI Act, which requires AI systems deemed “high-risk” to conduct fundamental rights impact assessments, and track risk management practices and incidents. Concerns have been raised over how a mandated “kill switch” could be verified by authorities, which is another area that is yet to be regulated according to our benchmark.
Between 2021 and 2024, an international trend towards regulating AI emerged, but more recently, this has fallen away, with many countries adopting a laissez-faire approach in the hopes of better enabling AI-led innovation and economic growth. According to our benchmark, there were no new proposals to regulate AI in 2025, as opposed to six in 2024 (Australia, Chile, Mexico, South Korea, Taiwan, Vietnam). In 2025, two countries (Australia and the US) formally abandoned their prior plans to regulate the technology. These abandonments were a notable shift from earlier international approaches on AI, which were defined by the EU’s AI Act and the 2023 Bletchley Park Declaration on AI safety. As our benchmark shows, this declaration was one of the last major international agreements on AI.
