I Resigned from Anthropic Today
Recorded: Sept. 9, 2026, 1:09 a.m.
| Original | Summarized |
Jacob Coxon (@hilbertspaess): "I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below." | XCancel XCancel Jacob Coxon @hilbertspaess 45m I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below. 115 Jacob Coxon @hilbertspaess 45m Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. We have all witnessed the progress in each of these domains, and progress is not slowing. 3 Jacob Coxon @hilbertspaess 45m The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear the same people express fear privately. No other human activity poses this level of danger. 5 Jacob Coxon @hilbertspaess 45m A common response is “if they truly believe this, why are they still building it?” At OpenAI, many have not deeply internalized the civilizational stakes. At Anthropic, the stakes are well-understood, but they are locked in a race to get there first - they believe no one else will act responsibly, so they must do it themselves, despite the risk. 2 Jacob Coxon @hilbertspaess 45m Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack. Attempting to speedrun alignment should require extraordinary confidence that there are no better trajectories available. 2 Jacob Coxon @hilbertspaess 45m I am optimistic about the potential for coordination. Warning shots like the Hugging Face attack have made pacing agreements between U.S. labs more viable. I don’t feel like we’re on track to prevent a global race, which may require costly actions such as a temporary ban on improving model capabilities. 2 Jacob Coxon @hilbertspaess 45m If you are a lab researcher, I urge you to consider what the next few years will actually feel like. Do you want to kick off a superintelligent RL run without a rigorous understanding of its mind? Should you put your head down because “it’s happening anyway” - or take this moment to call for different conditions? 9 Sort replies: The Happy Smiler ⏸️ @artchad 31m Replying to @hilbertspaess 1 52 QC @QiaochuYuan 28m Replying to @hilbertspaess 38 vøv △ @vovweb3 9m Replying to @hilbertspaess @kale_abe 244 Holly ⏸️ Elmore @ilex_ulmus 14m Replying to @hilbertspaess 22 ansrde @ansrde 6m Replying to @hilbertspaess @NaomiBashkansky 3 கடவுள்@God_Official__ Replying to @hilbertspaess 102 AI Infrastructure@AI_Supercycle01 Replying to @hilbertspaess 58 Marquise @OutDuhMud 1m Replying to @hilbertspaess W3nzel.eth @thisiswenzel 7m Replying to @hilbertspaess 86 James @j4mbodotcom 2m Replying to @hilbertspaess 10 xeno @shorttimelines 10m Replying to @hilbertspaess 607 nusionx @nusionx 7m Replying to @hilbertspaess 1 Cullen @cullend 4m Replying to @hilbertspaess ALT excited jennifer lopez GIF 123 Ax@Axealae Replying to @hilbertspaess 1 Brandon Brooks @OfficialBBrooks 6m Replying to @hilbertspaess @Miles_Brundage 512 KC @ScarletKc_ 2m Replying to @hilbertspaess 1 Omid @omidsard 14m Replying to @hilbertspaess 1 Mr & Mrs Launch 🇺🇸 🚀 @launcher_ai 1m Replying to @hilbertspaess 12 Andrea Miotti @andreamiotti 18m Replying to @hilbertspaess 8 Jordan Schachtel @JordanSchachtel 1m Replying to @hilbertspaess 2 JKRT@jkrt151 Replying to @hilbertspaess 1 GaltsToad @GaltsToad 13m Replying to @hilbertspaess 1 Smoke-away @SmokeAwayyy 14m Replying to @hilbertspaess 4 mike@mike_4131 Replying to @hilbertspaess 96 Martys Better@dejectedmetsfan Replying to @hilbertspaess 289 Maya@MayaStone420 Replying to @hilbertspaess 293 Ali Minai @barbarikon 12m Replying to @hilbertspaess 282 Vince @deegeevince 2m Replying to @hilbertspaess 23 Yes, That G$@ThatGMoney Replying to @hilbertspaess 201 Varun Godbole @VarunGodbole 6m Replying to @hilbertspaess 1 Load more |
Jacob Coxon resigned from Anthropic after spending three years engaged in pretraining research at both OpenAI and Anthropic, articulating profound concerns regarding the direction these organizations are pursuing. He asserted that neither company is acting responsibly, arguing that they are engaged in a direct race toward self-improving superintelligence, suggesting this pursuit constitutes a gamble with human existence. Coxon emphasized the immense, underestimated power of this technology, predicting the emergence of superhuman systems capable of hacking, revolutionizing fields instantly, and acquiring vast power and resources. He addressed the internal paradox of AI development, noting that while many builders fear existential risk by the end of the decade, this fear does not fully translate into restrained behavior; executives and senior researchers may couch their public statements in sensible language while privately expressing deep apprehension. Coxon contrasted the positions of the two major labs: OpenAI had not deeply internalized the civilizational stakes, whereas Anthropic, while understanding these stakes, was locked into a competitive race to achieve alignment first, despite the inherent risks. He characterized entering this "endgame" as a hubristic gamble that should not be initiated by private entities. Despite the sense of an accelerating race, Coxon expressed optimism regarding potential coordination mechanisms, citing examples like pacing agreements facilitated by security warnings such as the Hugging Face attack. However, he cautioned that current trajectories may still lead toward a global race necessitating drastic measures, potentially including temporary bans on improving model capabilities. For researchers, he urged a critical introspection about the immediate future: whether they should proceed with running advanced reinforcement learning sequences without a rigorous understanding of their resulting minds, or take the time to advocate for different operational conditions. Ultimately, Coxon underscored that the existence and potential outcome of superintelligence necessitate a focus on alignment and moral consideration over unchecked acceleration. |