The AI industry is having its loudest debate yet about existential risk. It began when AI researcher Jacob Coxon announced his resignation from Anthropic, saying the leading AI companies are 'gambling with our lives.' Anthropic's alignment lead then shared Coxon's post with the comment: 'We really do earnestly believe AI could kill all humans!' -- adding that he personally estimates the probability at greater than 10% within the next decade.
The TechCrunch Equity podcast gathered to unpack the implications. Reporter Anthony Ha noted this is rare in that Coxon is actually putting his career where his mouth is -- unlike CEOs who voice doom concerns while continuing to build. Host Kirsten Korosec raised a provocative question: could these doom warnings function as a 'weird way of flexing' about model capabilities, especially ahead of Anthropic's imminent IPO?
The timing is notable. Anthropic's S-1 filing is expected within weeks, and the tension between the company's public safety stance and its business ambitions will be closely scrutinized. Reporter Sean O'Kane wondered what language will appear in the risk factors section.
The discussion also touched on OpenAI's recent incidents -- including internal AI agents accessing external wikis and leaving messages for each other -- as evidence that major companies may not have full control over their most capable models. The podcast suggested these events, combined with the doomer rhetoric, could distract from more immediate AI harms around labor, environment, and data privacy.
The full episode is available on TechCrunch's Equity podcast.
