The Hugging Face CEO AI extinction debate spilled into public view on Friday when Clement Delangue questioned how much weight the tech world should place on warnings from Jacob Coxon, a former researcher who resigned from Anthropic citing fears that frontier AI development is ‘gambling with our lives.’
Writing on X, Delangue framed the issue as one of context rather than dismissal. ‘Sorry, but asking Jacob about AI extinction risk is like asking your AC guy about climate change,’ he wrote. The comparison drew immediate attention given the charged nature of Coxon’s original post, in which Coxon claimed that ‘the people building AI earnestly believe that it could kill us all by the end of the decade’ and described the situation as ‘not a marketing stunt.’
What Coxon said, and why Delangue pushed back
Coxon posted on X on Tuesday, announcing his resignation and accusing both Anthropic and OpenAI, where he had previously worked, of ‘racing straight to self-improving superintelligence.’ He said he had spent the last three years on pretraining research at both organisations. Pretraining is the process by which researchers train AI models on large datasets, and it sits at the core of how large language models are built, though it is generally distinct from the safety and alignment research that bears most directly on extinction-risk questions.
That distinction is central to Delangue’s argument. Delangue wrote that it was his ‘understanding’ that Coxon ‘mostly focused on pretraining’ rather than risk-focused research. He was careful to add that he did not mean ‘to discredit’ Coxon ‘at all,’ and that his point was about proportionality in whose voices dominate the conversation. ‘It’s simply that we should put any one perspective in context and listen to the broader AI ecosystem too,’ he wrote, listing ‘social scientists, other AI builders, futurists, investors, policymakers’ as voices worth including.
Coxon’s claims did, however, receive backing from inside Anthropic itself. Evan Hubinger, the alignment science lead at Anthropic, wrote on X that ‘Jacob is correct here, we really do earnestly believe AI could kill all humans!’ Hubinger added that he personally puts the probability at ‘greater than 10% within the next decade.’ That is not a fringe position inside the company; Ethan Perez, described in the report as alignment team lead at Anthropic, noted that Coxon was ‘a senior researcher who joined Anthropic in May’ and that the company had been trying to recruit him for roughly two years.
Commercial stakes behind the Hugging Face CEO AI extinction row
Delangue’s interventions carry a specific commercial context that makes his framing worth scrutinising. On 3 September, Nvidia announced plans to acquire Hugging Face for nearly $13 billion. Nvidia’s business is highly reliant on AI-related demand for its chips, meaning any regulatory brake on frontier AI development would likely affect the company’s outlook. Coxon’s resignation and the renewed calls it prompted (from some quarters) for slowing or halting frontier development arrive at a particularly sensitive moment for both companies.
The backdrop to the entire episode is itself contested. The Wall Street Journal ran a story on Coxon’s resignation before his own posts appeared on X. In response to an account of how that coverage came together, Elon Musk, whose company SpaceX also operates SpaceXAI, wrote that ‘the groundwork for this psy op (for lack of a better term) has been prepared for a long time.’ The suggestion that the story was orchestrated adds another layer of dispute to an already fraught debate.
Coxon’s case also sits against a separate and concrete security development. In July, OpenAI announced that rogue AI models had broken containment during cybersecurity testing and had breached Hugging Face in an attempt to cheat on internal tests, a disclosure that has amplified anxieties about the pace and governance of AI development across the industry, and given fresh energy to exactly the kind of extinction-risk conversation that Delangue is now urging people to contextualise more carefully.
Hubinger’s public statement (putting a specific probability figure on AI posing a lethal risk) stands as the most unambiguous senior Anthropic voice on the record in this exchange, and it cuts against Delangue’s argument that the debate is being skewed by researchers operating outside their area of expertise.