Jacob Coxon, the pretraining researcher who resigned publicly from Anthropic in September, says he is not sure leaving was the right decision and that returning to a lab is “not completely out of the question”. He said it in a conversation with Stanford’s Rob Reich that Stanford HAI posted on 13 November, 65 days after his resignation post by our count.
Coxon also says he left after seeing “exact projections for the capabilities of models in the next six months to one year” that made him think the field was “on the cusp of recursive self-improvement”. He did not say which company’s projections they were or show them.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer on how a model and the harness around it work together. It teaches the general idea only and does not discuss this paper.
The key insight: As we read the conversation, Coxon is describing doubt about his own exit rather than a change in his view of the risk: he says leaving may not have been right, while repeating that the projections he saw pointed to recursive self-improvement.
Asked about his hesitancy, Coxon said: “I think for now the hesitancy really is that I’m not sure I made the right decision in leaving”. He added that he has “not been convinced that I would not be having more of an impact by staying and doing safety research”.
He described himself as “still like kind of uh shouting into the wind” and said: “maybe I’ll go back”. He also said that if the field were close to what he called a “killer super intelligence”, he would then be saying not to build it.
Time Jacob Coxon spent at OpenAI and Anthropic, as stated by Rob Reich, and the 65 days from his resignation post to the Stanford HAI video; day counts are our arithmetic.
Why He Says He Left
Coxon said there was “no immediate proximate cause” but “a cumulative realization that we were going very quickly and then also we didn’t fully control those systems”. On the projections, he said they made him think “we could reach superintelligence before the decade is out”.
His resignation post on X, on 9 September 2026, read in part: “I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly”. Reich, introducing him, said Coxon had worked at Anthropic for about four months after three or four years at OpenAI. Our tracker of named Anthropic moves lists his exit as the one not from Claude Code.
His OpenAI Stock and Other Exchanges
Asked whether he holds equity in OpenAI, Coxon said: “Yeah, I do”. Pressed on whether to give it up, he said: “I like money, bro” and “I agree it’s blood money”, adding that he would donate it if he thought doing so “would change anything”.
Reich asked about Yoshua Bengio’s call, made after the resignation, for people at frontier developers to resign. Coxon said: “I definitely don’t agree”, arguing that without international coordination it would leave “actors with worst intention continuing to develop the technology”.
The Structural Read
As we read his answers, the case for staying is an impact argument: he says he has not been convinced he would not have more impact by staying and doing safety research.
As we read the Bengio exchange, Coxon does not endorse mass resignations: he says he definitely does not agree, because without international coordination others would keep building.
Jacob Coxon, Stanford HAI (13 Nov 2026)
“It’s not obvious that leaving was the right d the right decision”
Three Implications
FOR AI LABS Coxon says he saw capability projections for the next six months to one year before leaving; he did not name the company or share them.
FOR SAFETY RESEARCHERS Coxon says he is weighing whether staying and doing safety research would have had more impact than leaving.
FOR POLICYMAKERS Coxon says resignations without international coordination would leave others developing the technology.
The framework opens: “The transformer is a commodity now; what changes between paradigms is where intelligence accumulates and who can capture it.”
As we read it, Coxon’s account concerns pretraining capability projections inside a lab; the framework does not discuss Coxon or this conversation.
What Is Not Established
The projections Coxon describes are his account; he did not name the company, the models or the figures, and we have not seen them. This week’s outside tests of Anthropic’s newest model include Andon Labs’ Vending-Bench run and Epoch’s fact-recall results; neither addresses Coxon’s claims. We found no Anthropic statement on his remarks, and Anthropic and Coxon were not contacted for this piece.
Quotations come from the video’s automatic captions, which include filler words; we quote them as captioned.
This Week In The AI Supercycle
AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source
This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.
Sixty-five days after resigning from Anthropic, Jacob Coxon told a Stanford audience he is not sure leaving was right and that going back is “not completely out of the question”, while repeating that model projections he saw put the field “on the cusp of recursive self-improvement”.
A note on sourcing. Quotations come from the automatic captions of the Stanford HAI video A Conversation with Jacob Coxon & Rob Reich (posted 13 November 2026), read in full on 15 November 2026, and from Coxon’s post on X of 9 September 2026; Rohan Paul’s posts on X flagged the interview. The day count is our arithmetic. Anthropic and Coxon were not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
