Anthropic co-founder Jack Clark told BBC that society could eventually establish rules for an “emergency shutdown button” that a third party could use to completely disable artificial intelligence if it becomes dangerous. Clark said companies, including Anthropic, have various shutdown methods and called for the need for and oversight of such a mechanism to be included in broader policy discussions. Clark, who did not assign a percentage to the likelihood of human extinction, said an unregulated industry was a bad idea and that “enormous risks are being gambled with.” Jacob Coxon, who worked at OpenAI and Anthropic over the past three years, resigned after accusing the companies of irresponsibility; he argued that the systems could “hack everything” in the near future and that some employees believed they could “kill humanity” within 10 years.
Why it matters
The proposal opens up discussion of an externally operated oversight mechanism for artificial intelligence safety that is independent of companies’ own internal safeguards. Such regulation raises questions about at what risk threshold a system would be shut down, who would make that decision and under whose supervision. Clark’s objection to leaving the industry to regulate itself brings the question of whether responsibility for safety should rest solely with developer companies onto the broader policy agenda. Coxon’s resignation and his claims concerning cyberattacks and humanity’s existence provide grounds for questioning the relationship between companies’ risk assessments and the safeguards disclosed to the public. However, it remains unclear how the third party’s authority would be defined, how it would differ from existing shutdown methods and how the claims would be verified.
Background
Anthropic is not a new name in the FikirPilot archive: over the past 90 days, we published 13 articles mentioning the name; the most recent was dated September 15, 2026.