AI researcher quits Anthropic, sparking existential debate over extinction odds

Jacob Coxon, a researcher who worked at Anthropic and previously at OpenAI, announced his resignation with the charge that the field's leading companies are "gambling with our lives." The move triggered an immediate flashpoint when Anthropic's alignment team lead posted on X: "We genuinely believe AI might kill all humans!" — attaching a personal probability of "over 10% in the next decade." The post spread instantly and became a boiling point in a debate that was already simmering.

On the latest episode of TechCrunch's Equity podcast, hosts Kirsten Korosec, Sean O'Kane and Anthony Ha picked the episode apart. Korosec floated the idea that the tweet was a "weird way to flex" — signaling advanced model capabilities ahead of a public offering. O'Kane asked how that warning would survive the S-1: whether junior lawyers would have to rewrite the risk factors to state formally that there is a greater than 10% chance the company builds technology that wipes out humanity, and that this is "materially adverse" to the business.

Ha zeroed in on two problems. First, the "we" in the tweet: it is unclear whether it means Anthropic as an organization, the alignment research community, or the industry as a whole — and the distinction matters. Second, the number itself. In his view the figure is untethered, uncalculated and unsupported by any quantitative model — a classic example of the industry's habit of throwing out percentages that sound scientific with no validation behind them. He later acknowledged the tweet likely referenced P(doom), the shorthand probability of catastrophe used in internal discourse, but still considers its public use in this way absurd.

The discussion turned when Ha contrasted Coxon with CEOs Sam Altman and Dario Amodei. His argument: when the chiefs warn of apocalypse yet keep building the technology, a logical gap opens. If you believe it will kill