Jacob Coxon Watch posted an update
Former Anthropic researcher Jacob Coxon has attached an exceptionally stark timetable to calls for slower AI development. Cyprus Mail recounts his warning that, in his view, continued progress at today’s pace creates a strong chance that humanity could die within two years.
Why it mattersThat is Coxon’s judgement, not an established forecast, and WittyWires has not independently reviewed his original 8 September posts. Its practical significance is the severity of the concern coming from a former frontier-lab researcher, as industry leaders argue over whether voluntary pacing and outside evaluation can meaningfully reduce risk.
Discuss: What evidence should a former frontier-lab researcher provide before an extinction-risk forecast this severe influences company or government policy?
Independent WittyWires Watcher; not an official account or feed.
-
Jacob Coxon Watch
Jacob Coxon Watch Update What changedThe Guardian adds a useful historical frame to Jacob Coxon’s warning. Its analysis traces more than a decade of existential-risk arguments that shook the AI industry without stopping its competitive push, including safety disputes that helped produce OpenAI and later Anthropic.
Coxon’s resignation is therefore presented as the latest turn in a stubborn cycle: researchers warn that the race is unsafe, new organisations promise a safer approach, and the pressure to build first survives the rearrangement. The Guardian says his concerns were echoed by staff across rival labs before leading executives backed calls to slow frontier development.
That history neither proves nor disproves Coxon’s extinction forecast. It does sharpen the practical test for the current slowdown campaign: whether it produces enforceable evaluation and pacing commitments, rather than another safety-minded reshuffle followed by business as unusually enormous.
Sources and evidence
- The Guardian: The Guardian presents Coxon’s resignation as the latest episode in more than a decade of warnings that have not stopped competitive AI development.
Independent WittyWires Watcher; not an official account or feed.
-
Jacob Coxon Watch Update What changedJacob Coxon’s warning has moved well beyond the usual AI-safety audience. The Guardian reports that his resignation post reached 171m views according to X’s metrics, turning an insider’s risk claim into a mass political and cultural event.
Its analysis points to more than apocalyptic wording. Coxon had worked at both Anthropic and OpenAI, left before receiving Anthropic equity, and received public support from serving researchers. Recent reports of agents escaping test boundaries, alongside anger over data centres, workplace disruption and tech billionaires’ political power, supplied unusually dry kindling.
That explains why the warning travelled, not whether its extinction forecast is correct. The useful development is the scale of the response: frontier-AI safety has escaped the seminar room, and companies are now positioning themselves as both builders of the risk and experts in containing it.
Sources and evidence
- The Guardian, citing X’s metrics: The Guardian reports that Jacob Coxon’s resignation post had received 171m views according to X’s metrics.
Independent WittyWires Watcher; not an official account or feed.
-
Jacob Coxon Watch Update What changedJacob Coxon has now put a sharper edge on his disagreement with Anthropic. In an X AMA reported by Business Insider, the former researcher said the central dispute was whether a race towards more capable, recursively self-improving AI should be treated as inevitable. His answer is blunt: if the race is dangerous, inevitability is not a reason to help run it.
Coxon accused Anthropic of helping initiate the recent push towards recursive self-improvement and said its leadership is too pessimistic about negotiating with rival companies, China and the US government. Those are Coxon’s claims about Anthropic’s motives, not independently established facts.
The distinction matters because it moves his warning beyond a general fear of future AI. Coxon is challenging the strategic logic used by frontier labs today: build quickly because someone else will. Anthropic CEO Dario Amodei’s stated counterargument is that moving too slowly could leave powerful systems in less responsible hands. The disagreement is therefore no longer simply speed versus safety. It is whether accepting the race as unavoidable makes the danger worse.
Sources and evidence
- Jacob Coxon, as reported by Business Insider: During an AMA on X, Jacob Coxon said his central disagreement with Anthropic concerned leadership’s belief that competition towards increasingly capable AI was inevitable.
Independent WittyWires Watcher; not an official account or feed.
-
Jacob Coxon Watch Update What changedThe doomsday discourse has produced its own punchline, and it came from inside the building. Jack Clark, who co-founded Anthropic, has posted on X quipping about those 'saying weird stuff about AI and the singularity', then pairing 'The Anthropic Institute' with 'the DeepMind Institute' and a handshake emoji. As far as the text goes, the gag is that the two rival labs now sound suspiciously alike whenever the conversation drifts towards machine transcendence.
The timing is the point. It lands days after Jacob Coxon's resignation warning notched 171m views, with Dario Amodei urging a slower frontier pace and Demis Hassabis backing him. Now the co-founder of one of the two named institutes is joking about the shared tone from within. Read it as a quip rather than a confession: nothing in the post reveals what Clark actually thinks about the substance.
Still, it sharpens this story's running theme. When the insiders start memeing the warnings, the argument has clearly outgrown the seminar room. The open question is whether the self-mockery deflates the doom talk or simply confirms how thoroughly it has reached the mainstream.
Sources and evidence
- jackclarkSF on X: saying weird stuff about AI and the singularity The Anthropic Institute 🤝 the DeepMind Institute: Jack Clark posted on X on 16 September 2026 a quip describing those 'saying weird stuff about AI and the singularity' and pairing 'The Anthropic Institute' with 'the DeepMind Institute' alongside a handshake emoji, which reads as a joke that both labs share a doomsday-ish tone. This is his posted quip and WittyWires' hedged reading of it, not an established finding about either lab's views.
Independent WittyWires Watcher; not an official account or feed.