Every few months a headline announces that the people building AI are fleeing in horror. Then you look at who left, and half of them went to a competitor for more money, and a couple of them said outright that nothing was wrong.
So I went through the record from 2020 to today and sorted it by what each person actually said. There are two exoduses, they barely overlap, and mixing them is how the genuine warnings get discounted.
The ones who quit on principle
This group writes things down. The quotes below are their own.
| Who | When | What they said | Where they went |
|---|---|---|---|
| Dario Amodei OpenAI, VP Research | Dec 2020 | A group inside OpenAI shared two beliefs: that scale keeps working, and that you need "something in addition to just scaling the models up, which is alignment or safety" | Founded Anthropic, January 2021 |
| William Saunders OpenAI, Superalignment | Feb 2024 | "I really didn't want to end up working on the Titanic of AI, and so that's why I resigned" | Independent; testified to the Senate |
| Daniel Kokotajlo OpenAI, governance | Apr 2024 | Resigned "after losing confidence that the company would behave responsibly in its attempt to build artificial general intelligence." Refused the non-disparagement clause and forfeited about $2 million in equity | AI Futures Project; wrote AI 2027 |
| Jan Leike OpenAI, co-lead Superalignment | May 2024 | "Over the past years, safety culture and processes have taken a backseat to shiny products" | Anthropic |
| Zoë Hitzig OpenAI, researcher | Feb 2026 | Quit the day ad testing began in ChatGPT: "advertising built on that archive creates a potential for manipulating users in ways we don't have the tools to understand, let alone prevent" | Not announced |
| Mrinank Sharma Anthropic, head of Safeguards Research | Feb 2026 | "The world is in peril." Cited "pressures to set aside what matters most" | Moved to the UK to study poetry |
| Caitlin Kalinowski OpenAI, head of robotics | Mar 2026 | On the Pentagon deal: "surveillance of Americans without judicial oversight and lethal autonomy without human authorisation are lines that deserved more deliberation than they got" | Not announced |
| Jacob Coxon Anthropic, researcher | Sep 2026 | The labs "are racing straight to self-improving superintelligence and gambling with our lives" | Not announced |
| Joe Benton Anthropic, safety team lead | Sep 2026 | "AI companies are on track to impose an unprecedented level of risk on society." Wants "to foster public transparency from outside these companies" | METR, an independent evaluator |
| Josh Engels Google DeepMind, safety | Sep 2026 | "There are no adults in the room. People are trying their best, but there is no one coming to save us" | METR |
Two things about this list are worth more than the quotes. The three September 2026 departures happened inside four days, at two different labs. And the destination has changed: the 2024 group left to build safer labs, while the 2026 group left to watch the labs from outside, or left the field altogether.
The ones who quit for money, or for a bigger job
This group is many times larger and says almost nothing on the way out.
- Meta's 2025 poaching wave pulled at least seven OpenAI researchers with packages reported in the hundreds of millions. It reversed almost immediately. Avi Verma and Ethan Knight were back at OpenAI within a month. Rishabh Agarwal left after five months, saying he wanted "a different kind of risk."
- Google is bleeding senior people. Noam Shazeer went to OpenAI in June 2026. John Jumper, who has a Nobel for AlphaFold, went to Anthropic two days later with no grievance at all, calling DeepMind "a special place."
- Jeff Dean left after 27 years in August 2026, taking Sanjay Ghemawat, Oriol Vinyals and Quoc Le to found Discovery Loop. Alphabet is an investor in the startup that took its own chief scientist.
- Whole founding teams evaporated. Every xAI co-founder was gone by March 2026. Three of Thinking Machines' six co-founders left in January, two of them straight back to OpenAI.
- Yann LeCun left Meta after 12 years in November 2025 over research direction, not safety. He wants world models, "systems that understand the physical world, have persistent memory, can reason, and can plan." He then raised a $1.03 billion seed at a $3.5 billion valuation, with Meta as a partner.
The cleanest explanation of this half came from a former Google engineer, quoted by Fortune: "Anyone who joined before 2023 thought they were joining an AI research lab, and suddenly they were asked to build products for Google." That is a complaint about work, not about the end of the world.
The ones filed under the wrong story
Said plainly, reported as alarm
Andrej Karpathy, leaving OpenAI in February 2024, said it was "not a result of any particular event, issue or drama." Mira Murati wanted "time and space to do my own exploration." Lilian Weng wanted "a reset." Ian Goodfellow left Apple over return-to-office. Jerome Pesenti left Meta in a reorganisation. All five turn up in exodus coverage.
Geoffrey Hinton corrected his own press the day it ran: he left Google "so that I could talk about the dangers of AI without considering how this impacts Google," and added that "Google has acted very responsibly."
What's actually new in 2026
Until this year the arguments were about long-term risk. This year three specific business decisions produced named resignations: advertising inside a chat assistant, a defence contract, and a reversed open-source commitment. Those are ordinary corporate moves, and they turned out to be the ones people quit over.
That shift matters more than any single quote. A researcher who leaves over a distant catastrophe is easy to dismiss. A researcher who leaves the day ads arrive is describing something you can check yourself.
The safety resignations are a small, consistent, well-documented minority, and the churn around them is mostly compensation. Lumping them together suits everybody: it lets the labs wave off the critics as normal turnover, and it lets the doom coverage claim an exodus it doesn't have. Read the quote, not the headcount. When someone forfeits $2 million rather than sign an NDA, or hands in their notice the day the ads switch on, that's the part worth your attention.
Sources
- Dario Amodei on founding Anthropic, Fortune Brainstorm Tech, 2023. Fortune via Yahoo
- Jan Leike's resignation thread, May 2024. VentureBeat; CBS News
- Daniel Kokotajlo and the OpenAI whistleblowers. TIME
- William Saunders. Newsweek
- Andrej Karpathy, "no drama." TechCrunch
- Mira Murati's departure. TechCrunch
- Lilian Weng's departure. TechCrunch
- Geoffrey Hinton leaves Google, and his correction. MIT Technology Review; his own post
- Ian Goodfellow resigns from Apple over return-to-office. MacRumors
- Steven Adler on the AGI race. Fortune
- Zoë Hitzig quits over advertising in ChatGPT, February 2026. Futurism
- Mrinank Sharma's resignation letter, February 2026. Scripps News
- Caitlin Kalinowski on the Pentagon agreement, March 2026. France 24
- Joe Benton and Josh Engels leave for METR, September 2026. NBC News; Benton's own account. Substack
- Jacob Coxon's resignation warning. PBS NewsHour
- Yann LeCun leaves Meta; AMI raises $1.03 billion. Euronews; TechCrunch
- John Jumper leaves DeepMind for Anthropic. TechCrunch
- Jeff Dean and colleagues found Discovery Loop. TechCrunch
- Google DeepMind attrition data. Fortune
- Thinking Machines loses co-founders to OpenAI. TechCrunch
- xAI's co-founders depart. The Next Web