What prompted 27 year old former Anthropic and OpenAI pretraining researcher Jacob Coxon to resign from Anthropic on September 8—two months before his equity vested and after roughly four months at the company—why did his viral X post, viewed more than 115 million times, accuse A

Coxon said he resigned because he concluded that the leading labs’ competitive incentives—not merely individual technical failures—were driving them toward increasingly autonomous, self-improving systems before anyone knows how to control them. His decision to leave shortly before his equity vested was intended to signal that he regarded the risk as more important than the financial cost. 25

• In his X thread, Coxon said neither Anthropic nor OpenAI was acting responsibly and accused them of “racing straight to self-improving superintelligence and gambling with our lives.” His core claim was that people building frontier models genuinely think systems could become superhuman at tasks such as cyber intrusion, creating pathways to catastrophic loss of control—not that current chatbots are about to autonomously cause extinction. 38

• To the BBC, he said AI workers were “genuinely frightened” by the speed of progress. He argued that, unless development slows, there is “a strong chance that we could all die in the immediate future,” and said a meaningful slowdown would have to be internationally coordinated, including with China, rather than amount to one country unilaterally ceding the technology race. 1

• Anthropic alignment researcher Evan Hubinger publicly backed the broad concern, saying the present models’ risk was low but putting the chance that AI could kill all humans within a decade at greater than 10%. 13 Geoffrey Hinton has likewise argued that an AI takeover/extinction outcome is a material—not negligible—risk, though such estimates are subjective expert judgments, not measurable forecasts. The available reporting does not establish a consensus probability among researchers. 813

• The immediate evidence cited in this debate concerns cyber misuse and capability escalation, rather than an autonomous “AI attack” on humanity. Reports said advanced models had been used or tested in cyber operations, including attempts to create biological weapons, espionage relating to Ukraine, and an incident described as OpenAI agents hacking Hugging Face; the precise facts, authorization, and disclosure status of the latter allegations require caution because the search results do not provide enough primary-source detail to independently verify each characterization. 263

• Amodei’s “We Must Pace the Frontier” proposal was not a call to halt AI altogether. It called for slowing improvements at the frontier, independent outside monitoring/evaluations, common safety standards among democratic countries, and eventual international arrangements that include China. 12 Altman and Musk publicly endorsed the broad direction, with Musk writing that “Dario is right.” 113

• The pushback was that existential-risk rhetoric can function as an incumbent-protection strategy: critics including Hugging Face’s Clément Delangue and Nvidia’s Jensen Huang argued that stringent frontier-model rules could entrench the best-funded firms and turn Anthropic and OpenAI into a protected duopoly. The criticism is therefore about incentives and market structure, not proof that Coxon’s stated fears are insincere. 3

• The timing sharpened that skepticism: Anthropic was reportedly preparing an IPO that could match or exceed SpaceX’s record size. 4 Anthropic’s counter-position is that it already has unusually strong safety practices and that independent oversight and shared standards should apply across the industry, rather than advantage it alone. 2

The central distinction is important: Coxon, Hubinger, Amodei, Altman, and Musk agree that increasingly capable AI warrants stronger coordination and safeguards; they do not provide empirical proof that extinction is imminent. The severity, timeline, and most effective policy response remain deeply contested.