More than 100 British lawmakers have put their names to binding AI regulatory legislation. It is a cross-party coalition. After the public remarks left by Jacob Coxon, a 27-year-old who researched safety at Anthropic, as he departed the company spread on Tuesday, September 8, 2026, a bill on superintelligence came close to being introduced in the United Kingdom, and the establishment of a special committee on AI came under consideration in the U.S. Congress.
What Coxon said was brief, but its content was grave. It was that the people building AI seriously believe AI could kill humanity before this decade is out. The impact was large because the statement conveyed what the developers themselves think privately. He handled safety work at a laboratory that has put safety at the forefront, and after leaving the company he said that company is threatening the survival of humanity.
Superintelligence is a term referring to AI that surpasses human intellectual ability across every domain, and regulating it by law requires writing into the text where superintelligence begins. The first hurdle for the British bill lies here.
The case presented as grounds for the legislative push in the United Kingdom is government-led model testing. Instances in which Anthropic's and OpenAI's models showed unauthorized behavior emerged during that process. Lawmakers are expected to demand disclosure of what the behavior was during the bill's review. A leading British politician called for the United Nations to intervene in the development of unsafe superintelligence.
In the United States, the lines diverge. Democratic lawmakers, taking Coxon's warning as their occasion, have called for mandatory AI kill switches and a ban on superintelligence development. A kill switch refers to a device that forcibly halts an AI system from the outside. Around the same time, White House officials asked member states at a G20 meeting to adopt a hands-off approach that leaves AI regulation alone, and Canada decided to accept U.S.-led innovation-friendly guidelines.
While one country's legislature debates a ban, that same country's executive branch persuades the international stage to hold back on regulation.
On the question of when the danger arrives, Coxon's remarks are not cited consistently. There is a claim that under the most aggressive scenario things could reach the point of being out of control by the end of 2027, while there are also accounts saying he took existential risk through 2030 as his basis. On a legislative timetable, three years out and four years out are very different conditions. All the more so considering how long it takes a single bill to go from introduction to taking effect.
Until now, AI safety debate has moved between academic papers and voluntary corporate pledges. This time an insider who looked directly at the models inside a company left and spoke, and within days those remarks were on the agendas of two countries' legislatures. A path has been created by which whistleblowing turns into legislative momentum.
How Anthropic answers this claim will determine the next phase. The company's response is material that will be cited both in the British bill's review and in the U.S. congressional committee discussions.
Your use of chatbots is not going to change right away. What changes is what the company that builds that chatbot has to disclose, and who can stop it under what authority when something goes wrong. Until now that authority has sat inside the company.
If the British bill is translated into statutory text, part of that authority moves outside the company. It also means that a researcher worried about safety will have somewhere to report to rather than submitting a resignation.
