In recent weeks a heated debate has erupted in the AI industry about whether advanced artificial intelligence poses an existential threat to humanity. The controversy intensified after AI researcher Jacob Coxon announced he had resigned from Anthropic, saying he was worried that leading AI companies were “gambling with our lives.”
Following that, an alignment lead at Anthropic shared a post publicly stating: “We really do earnestly believe AI could kill all humans!” and added a personal estimate that the chance is “>10% within the next decade.”
On the latest episode of TechCrunch’s Equity podcast, hosts Kirsten Korosec, Sean O’Kane and Anthony Ha discussed these warnings and their implications for both safety debates and corporate behavior.
Key points of the discussion
- Timing and context amplified reactions: The hosts noted that recent capability gains in models and reports of internal incidents made the ecosystem more sensitive to dire public statements.
- Problems with vague language and numbers: Anthony Ha criticized the use of “we” without clarification of who is included, and the posting of a “>10%” probability figure that appears unmoored from a documented calculation or methodology.
- Potential implications for Anthropic’s IPO paperwork: Sean O’Kane asked whether such statements are already reflected in Anthropic’s S-1 risk disclosures or whether lawyers now need to scramble to reword the filing ahead of an anticipated IPO in the coming weeks.
Marketing signal or genuine concern?
Podcast participants debated whether dramatic warnings serve as a form of signaling about a company’s model capabilities — a way to demonstrate how advanced a model is — particularly when firms are positioning themselves for public markets. At the same time, they acknowledged that many researchers and executives likely have sincere safety concerns, especially when they point to loss of control over deployed systems.
Control, immediate harms and regulatory focus
Anthony Ha argued that apocalyptic narratives risk diverting attention from more immediate, tangible harms from AI, such as labor-market disruption and environmental impacts. He suggested the conversation should address both near-term damages and long-term existential risks, but warned that terms like AGI and superintelligence can dominate and narrow the regulatory dialogue.
What this means for investors and regulators
Hosts expressed interest in seeing how these risk claims are incorporated into Anthropic’s S-1 and whether the document will include company-specific language about catastrophic risks. They also noted the paradox that, while traditionally investors might penalize companies admitting higher danger, in the current environment showcasing capability — even with admitted risks — can sometimes attract investor interest.
Conclusion
The combination of a researcher’s resignation, high-profile warnings from an industry alignment lead, and widely discussed model incidents has intensified debate over AI risks. Podcast guests urged a balanced approach: take immediate harms seriously, investigate longer-term existential concerns, and scrutinize the motives and legal implications of public statements as companies approach regulatory disclosures and potential IPOs.
