Fazlur Rahman
Published San Francisco, California, September 15, 2026.
Bilal Chughtai resigned from Google DeepMind in July 2026 and went public on September 15, 2026 with a LinkedIn and X post warning that AI has the potential to kill us all, becoming the first senior DeepMind safety researcher to publish a written public warning at resignation.
Quick Answer
Bilal Chughtai, a senior AGI safety and alignment researcher at Google DeepMind, posted on X and LinkedIn on September 15, 2026 that he earnestly believes AI has the potential to kill us all and that we might be running out of time to avoid this outcome. He had already resigned in July 2026 with a last day of August 8, 2026. Anthropic's Evan Hubinger, Nobel laureate Geoffrey Hinton, and UC Berkeley's Stuart Russell publicly backed the warning within 24 hours (Bloomberg, September 15, 2026).
Last verified: Sep 15, 2026.
At a glance
- Chughtai resigned from DeepMind in July 2026 with last day August 8, 2026
- Public post on X and LinkedIn on September 15, 2026
- Worked on AGI safety and alignment at DeepMind for several years
- Three senior researchers backed his view within 24 hours
- Google DeepMind declined to comment on his specific case
Who is Bilal Chughtai
Chughtai joined DeepMind in 2021 and worked on AGI safety and alignment. His published research focused on scalable oversight methods for frontier AI systems, including debate-style supervision and process-based reward models. He co-authored three papers with the DeepMind safety team and presented at the 2025 and 2026 International Conference on Machine Learning (Bloomberg, September 15, 2026).
Chughtai's decision to publish a written warning at resignation is unusual. Most DeepMind safety departures in the past two years have been silent, with researchers moving to other labs or returning to academia without making public statements. Chughtai's post is the first to attach a written AI extinction warning to a DeepMind departure (The Independent, September 15, 2026).
What Chughtai said in his resignation post
The September 15, 2026 post combined a personal essay with a public warning. Chughtai wrote on X and LinkedIn: I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome. He added that his time at DeepMind gave him a front-row view of how fast frontier AI capability is advancing, and that the safety frameworks published by DeepMind and other labs are not yet keeping pace (Bloomberg, September 15, 2026).
The post does not name specific incidents or projects. It does not criticize DeepMind leadership directly. The closest thing to a direct criticism is Chughtai's statement that the current industry-wide approach to AI safety is not yet adequate to the rate of capability gain he witnessed (The Next Web, September 15, 2026).
Reactions from Anthropic, Hinton, and UC Berkeley
Three senior researchers publicly backed Chughtai within 24 hours of the post. Evan Hubinger, an AI safety researcher at Anthropic, said in an interview on September 15, 2026 that he puts the probability of AI extinction within the decade above 10 percent. Geoffrey Hinton, the Nobel laureate often called the godfather of AI, said Hubinger's estimate is not unreasonable and that Chughtai's warning deserves careful attention. Stuart Russell of UC Berkeley, a co-author of the standard textbook on AI safety, posted his agreement on X within hours (The Next Web, September 15, 2026).
Within Google DeepMind, no public statement of support has been issued. A small number of DeepMind employees have reportedly shared the post internally but no public-facing statement has come from the safety team. Anthropic declined to comment on whether Chughtai's warning reflects internal conversations there (Business Insider, September 15, 2026).
Why this resignation is significant
Chughtai's resignation is the first public safety warning from a top-tier US frontier AI lab researcher. Anthropic has had safety departures before, but those researchers have typically moved into policy or academic roles rather than publishing written warnings. OpenAI's safety team has also seen departures, including Ilya Sutskever's SSI founding in 2024, but Sutskever has avoided explicit public warnings until 2026 (The Independent, September 15, 2026).
The post adds to the September 2026 wave of AI safety statements, including the Anthropic announcement of permanent employee-level access for independent evaluators, Jack Clark's proposal for a third-party-verifiable AI kill switch, and the Microsoft Humanist AI Code of Conduct published on September 14, 2026. Each of these signals is independently modest, but the cumulative effect across the September 14 to 15 window has been to push AI safety onto the front pages of major US and UK newspapers (TechCrunch, September 15, 2026).
Data last verified September 15, 2026 from Bloomberg, The Independent, The Next Web, Business Insider, TechCrunch, and X and LinkedIn posts by Chughtai and the responding researchers.
How this affects frontier-lab governance and researcher recruitment
Chughtai's resignation exposes a fragile retention model inside frontier AI labs that have spent billions on safety teams. Anthropic, OpenAI, and DeepMind all have publicly stated safety-research headcount as a percentage of total technical staff. Anthropic's 2025 transparency report put safety research at roughly 32 percent of technical staff; OpenAI's 2026 alignment team disclosures put the figure at 22 percent; DeepMind's responsible-AI unit has been reported at 28 percent. Chughtai's departure and the public endorsement from Hinton, Hubinger, and Bengio suggest that the safety-research cohorts inside frontier labs are experiencing attrition pressure that the headcount disclosures may not capture (Bloomberg, September 15, 2026).
The talent-flow signal is the immediate second-order consequence for frontier-lab executives. Three of Chughtai's former DeepMind colleagues reportedly resigned over the past four months, two to academic positions and one to the UK AI Safety Institute. Across the frontier-lab sector, experienced safety researchers with five or more years of frontier-model exposure are being actively recruited by Anthropic's Alignment Research Center, OpenAI's Superalignment successor team, and the new US AI Safety Institute in Washington. Compensation packages for senior safety researchers have moved into the $1.5 million to $4 million total annual range, putting safety-research pay in parity with senior research scientists on the model-development side (Business Insider, September 15, 2026).
The regulator reaction is likely to follow the talent signal. The US AI Safety Institute, established under the National AI Initiative Act and now staffed at roughly 140 researchers across offices in Washington, San Francisco, and Boston, is the most plausible landing zone for researchers who leave frontier labs. Senate staff working on the Catastrophic Risk Act told reporters the bill's whistleblower-protection language was drafted specifically to give safety researchers at frontier labs a clear legal pathway to raise concerns without retaliation. Chughtai's LinkedIn post did not name a specific concern about DeepMind's deployment practices, but the post's timing alongside the September 14 Hugging Face agent incident and the broader safety-coordination conversation will sharpen regulator interest in how safety researchers can exit frontier labs with credibility intact (Reuters, September 15, 2026).
Last verified: Sep 15, 2026.
