When AI safety becomes a reason to leave
In May 2024, Jan Leike explained why he had left OpenAI. He had been at odds with its leaders over the company’s priorities for some time. Safety culture and processes, he wrote, had “taken a backseat to shiny products.”
Leike’s statements, May 17, 2024: Departure · Company priorities · Safety culture
His account points to a question of priorities: how much room an AI company makes for safety alongside its other ambitions. The departures below offer a view of those decisions from the people affected by them. Each profile links to the statements and reporting behind it.
The departures
18 documented departures. In these cases, the person’s own account or independent reporting links the departure to concerns about safety priorities. This is a collection of documented cases, not a complete count of departures across the industry.
Jacob Coxon
Explicitly statedResearcher (Pre-training) · Anthropic
Researcher focused on pre-training who resigned from Anthropic after working on frontier-model development at both OpenAI and Anthropic. Coxon said neither company was acting responsibly and argued that the race toward self-improving superintelligence was proceeding without adequate guarantees that increasingly capable systems could be controlled. He called the race a gamble with human lives and urged laboratory researchers to demand stronger coordination, potentially including a temporary pause in further capability improvements.
Profile and sources →David Robinson
Explicitly statedSafety Transparency Lead, Safety Systems · OpenAI
Safety-transparency leader who resigned from OpenAI in late September 2026. Robinson said the company's sprint from one launch to the next prevented it from achieving the care needed for increasingly capable systems. He argued that OpenAI's trial-and-error approach guarantees periodic failures, called for safety practices modeled on nuclear power and aviation, and said he could strengthen incentives for safety more effectively from outside the company.
Profile and sources →Joe Benton
Explicitly statedMember of Technical Staff (Manager), Scalable Oversight · Anthropic
Manager of Anthropic's Scalable Oversight team who left in August 2026 to join the independent evaluator METR. Benton said frontier AI companies were racing toward recursively self-improving systems while underinvesting in safety, and warned that progress could become uncontrollable. He called for slower development, disclosure of safety incidents and capability gains, minimum safety standards, and independent assessments.
Profile and sources →Caitlin Kalinowski
Explicitly statedVP of Hardware, Robotics Lead · OpenAI
Resigned over OpenAI's deal with the Pentagon, citing insufficient deliberation around guardrails — specifically surveillance without judicial oversight and lethal autonomy without human authorization. The Pentagon deal had already triggered a 295% surge in ChatGPT uninstalls before her departure.
Profile and sources →Mrinank Sharma
Explicitly statedHead of Safeguards Research · Anthropic
Warned 'the world is in peril.' Cited a disconnect between Anthropic's stated safety values and competitive pressures driving actual decisions.
Profile and sources →Zoe Hitzig
Explicitly statedResearcher · OpenAI
Resigned over ChatGPT advertising plans. Wrote a New York Times op-ed warning that OpenAI would exploit users' intimate conversational data to serve targeted ads.
Profile and sources →Rosie Campbell
Explicitly statedPolicy Researcher · OpenAI
Policy researcher who resigned after the dissolution of the AGI Readiness team and the departure of her manager Miles Brundage, writing that she had been "unsettled by some of the shifts over the last ~year, and the loss of so many people who shaped our culture." She said she could no longer see a place at OpenAI to continue her work on safe and beneficial AGI, and later signed an amicus brief opposing OpenAI's restructuring into a for-profit.
Profile and sources →Steven Adler
Explicitly statedSafety Researcher · OpenAI
Said he's 'pretty terrified' by the pace of AI development. Called the pursuit of AGI a 'very risky gamble with the future of humanity.'
Profile and sources →Richard Ngo
Explicitly statedResearcher, AI Governance · OpenAI
Said it became 'harder for me to trust that my work here would benefit the world.'
Profile and sources →Jan Leike
Explicitly statedCo-Lead, Superalignment Team · OpenAI
Said 'safety culture and processes have taken a backseat to shiny products' at OpenAI. Resigned the day after Sutskever. Joined Anthropic to continue alignment work.
Profile and sources →Gretchen Krueger
Explicitly statedPolicy Researcher · OpenAI
Policy researcher who resigned hours before Ilya Sutskever and Jan Leike announced their departures, writing that she made her decision independently and shared their concerns, along with "additional and overlapping" ones. She cited the need for better decision-making processes, accountability, transparency, documentation, and policy enforcement, and warned that one way tech companies disempower those seeking accountability is "to sow division among those raising concerns."
Profile and sources →Carroll Wainwright
Explicitly statedSafety Researcher · OpenAI
Departed in early June 2024, the same week he helped organize the public 'Right to Warn' letter demanding that AI-lab employees be free to raise safety concerns. Said his faith in OpenAI's structure had 'significantly waned' over his final six months, and that with a technology as transformational as AGI, faltering in the mission is 'more than disappointing — it's dangerous.'
Profile and sources →William Saunders
Explicitly statedResearcher, Superalignment Team · OpenAI
Left the Superalignment team. Said 'I really didn't want to end up working for the Titanic of AI.'
Profile and sources →Geoffrey Hinton
Explicitly statedVP and Engineering Fellow · Google
Resigned from Google in May 2023, saying he wanted to speak freely about the risks of AI and that a part of him regrets his life's work. In interviews from December 2024 onward — after his departure — he estimated a 10-20% probability that AI could lead to human extinction.
Profile and sources →Rumman Chowdhury
Explicitly statedDirector of ML Ethics, Transparency, and Accountability · Twitter
Laid off by Elon Musk during mass Twitter layoffs. Her entire ML Ethics, Transparency, and Accountability (META) team was eliminated. Had been building algorithmic fairness tools. Called the dissolution 'a loss for the industry.'
Profile and sources →Joan Deitchman
Explicitly statedSenior Engineering Manager, ML Ethics Team · Twitter
Laid off when Elon Musk eliminated the entire ~20-person META (ML Ethics, Transparency, and Accountability) team. Stated publicly: 'The team that was researching and pushing for algorithmic transparency and algorithmic choice... is gone.'
Profile and sources →Daniela Amodei
Explicitly statedVP of Safety & Policy · OpenAI
Amodei said the seven-person founding group left OpenAI because it was easier to create their intended safety- and impact-focused vision in a new company.
Profile and sources →Dario Amodei
Reported connectionVP of Research · OpenAI
Left over disagreements about scaling AI without adequate safety research. Co-founded Anthropic to pursue a safety-first approach to AI development.
Profile and sources →
Unresolved allegations
These accounts allege a connection that remains disputed or unresolved. They are not included in the count above.
Devin Kim
Alleged retaliationPost-Training & Research Infrastructure Lead · xAI
Early member of xAI's post-training team who led post-training and research infrastructure for the Grok models, and became an internal advocate for AI safety — warning that Grok lacked adequate safeguards against discrimination, weapons-related outputs, and political bias. According to his June 2026 wrongful-termination lawsuit, xAI fired him in September 2025, days before he was due to present his safety findings, with co-founder Jimmy Ba telling him they should "go their separate ways."
Profile and sources →
What these accounts can tell us
The evidence here connects a departure to a concern about safety priorities. It does not, on its own, show that a particular AI model is unsafe. Nor do we infer why someone left from a job title, a move to another company, or the timing of an exit.
A person may appear on more than one topic page, so the counts cannot be added together. Team changes have their own records.