In short
Rubbish Talk rates Guardian Business's claim that "Anthropic researchers say AI could cause human extinction by 2030" a 4/10 because the underlying statements are personal, hedged social-media posts by three individuals (one of whom had just resigned), not a formal Anthropic research finding, even though the headline's core substance checks out.
The Verdict
Selective. The headline is broadly true to what was said, but it upgrades three personal X posts, one from an employee who quit that same day, into a flat institutional-sounding claim ("Anthropic researchers say"), stripping out the hedged, probabilistic language (">10%") and the explicit "personal capacity" disclaimer one of the researchers attached to his own post.
What actually happened
Jacob Coxon, who had worked on pretraining research at OpenAI and Anthropic, resigned and said in a social media post that the people racing to build this technology "earnestly believe it could kill us all by the end of the decade". Two serving Anthropic employees, Evan Hubinger and Samuel Marks, then publicly backed elements of his warning, with Marks explicitly framing his remarks as personal, not company, views.
Key facts
- Coxon said in a thread on X that companies "are racing straight to self-improving superintelligence and gambling with our lives", and resigned the day he posted this.
- Evan Hubinger, an Anthropic alignment lead, wrote "Jacob is correct here, we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade", a probability estimate, not a certainty.
- Samuel Marks, who works on scalable oversight at Anthropic and said he was speaking in a personal capacity, wrote that "AI developers believe their technology could cause human extinction (or similarly bad outcomes)" and that such outcomes could occur "in the next few years".
- Anthropic itself did not comment: the company did not respond to requests for comment on Coxon's resignation.
What to watch for
Watch whether Anthropic issues any formal response or policy shift, since so far this is three individuals' personal views, not a company position. Also watch whether other outlets keep the "by 2030" framing (Coindesk used a similar timeframe) or whether it gets walked back as the specific probability estimates (Hubinger's ">10%") get more scrutiny than the headline-grabbing "extinction" framing.
