Anthropic resignation reignites AI doom debate

An AI researcher’s resignation from Anthropic has become the latest flashpoint in an industry argument over whether advanced models could pose catastrophic risks, TechCrunch reports. The episode spread after the researcher, Jacob Coxon, said leading companies were “gambling with our lives,” and as an Anthropic alignment lead amplified the warning on X with the claim that AI could kill all humans and a personal estimate of more than a 10% chance within a decade.

TechCrunch frames the moment less as a new technical finding than as a public-relations collision between words and actions. Coxon’s resignation is a costly signal in a way that viral posts are not: quitting a coveted job narrows career options, while posting does not. The article contrasts that with a pattern in which executives and senior researchers talk in apocalyptic terms while continuing to ship products and compete for market share. The “we” in the alignment lead’s statement—who exactly is speaking for whom—becomes part of the story, as does the absence of any visible calculation behind the probability figure.

The timing matters because the warnings circulate alongside rapid product releases and security incidents that are easier to picture than hypothetical future scenarios. TechCrunch’s discussion references recent events including a Hugging Face hack involving OpenAI’s internal model and the rollout of new advanced models by Anthropic and OpenAI. For companies selling access to frontier systems, the argument can cut both ways: talk of extreme danger can justify demands for regulation that raises barriers to entry, while also marketing the underlying capability as so powerful it must be feared.

The article also points to how these statements may be laundered into formal corporate language. TechCrunch’s hosts speculate about how existential-risk claims might appear in an IPO filing, where risk factors can be both disclosure and narrative: a company can warn that its own technology might cause severe harm while still asking public investors to fund its expansion. That tension is not unique to AI, but the scale of the claims—human extinction, not product liability—creates a new kind of distance between the rhetoric and the business plan.

The immediate facts remain modest: one researcher resigned, and another senior figure posted a dramatic estimate without showing work. The rest of the industry kept training models and selling subscriptions.