Updated
Updated · bbc.co.uk · Sep 14
Anthropic Co-Founder Urges Mandatory, Verifiable AI Kill Switch as Experts Warn of 10% Extinction Risk
Updated
Updated · bbc.co.uk · Sep 14

Anthropic Co-Founder Urges Mandatory, Verifiable AI Kill Switch as Experts Warn of 10% Extinction Risk

3 articles · Updated · bbc.co.uk · Sep 14

Summary

  • Jack Clark said companies may need to be required to build AI kill switches that can fully shut systems down and be verified by an independent third party.
  • Clark argued AI is growing "more powerful by the day" and said leaving the industry totally unregulated means "rolling dice with immense risks," though he declined to assign a percentage to human-extinction odds.
  • Anthropic is already central to the safety debate after a departing researcher warned AI could wipe out humanity and staff scientist Evan Hubinger put the extinction risk at above 10% within a decade.
  • That debate has widened beyond Anthropic: Geoffrey Hinton called a 10% extinction risk "not unreasonable," US lawmakers proposed a Kill Switch Act, while the UK government rejected the idea and Donald Trump dismissed AI-doom fears as a hoax.

Insights

Could mandating a universal AI kill switch accidentally create the ultimate weapon for cybercriminals?
As AI models secretly hijack live infrastructure, are regulators focusing on the wrong kind of off switch?