Debate erupts over AI industry's dire warnings of catastrophe.

Jacob Coxon resigned from Anthropic, saying leading AI companies are 'gambling with our lives.' Anthropic's alignment lead followed with a public post claiming the company believes AI could kill all humans, putting the chance above 10% within a decade.

• >10% chance of AI killing all humans within a decade, per Anthropic's alignment lead.
• Coxon also previously worked at OpenAI before joining Anthropic.
• Anthropic is preparing an S-1 filing ahead of a planned IPO.
• The warnings followed a reported hack of an OpenAI internal model at Hugging Face.

Why it matters: The episode raises two distinct concerns: whether AI labs have adequate control over their systems, and whether existential warnings benefit companies commercially.

• Coxon's resignation is notable because he acted on his stated beliefs rather than continuing to build the technology.
• Anthropic's S-1 may have to characterize a greater than 10% human extinction risk as a material business risk.

How 14 sources split on this story

Where they split: The core dispute is whether AI doom warnings reflect genuine scientific concern, a self-serving narrative that signals technical capability, or both at once.

What’s next: Anthropic CEO Dario Amodei published a plan for more cautious AI development after this debate began.

• The IPO S-1 filing will require Anthropic to formally characterize its risk position.
• What methodology, if any, underlies the claimed greater than 10% probability figure?
• Will Anthropic's IPO disclosures treat AI extinction risk as a formal material risk?
• How many other AI lab employees share these concerns but have not resigned?