Google DeepMind is losing researchers over its military work, and for the first time, two of them are naming the contracts that drove them out. London-based research engineer Sofia Liguori resigned this week and AI researcher Andreas Kirsch left earlier in October, both telling the BBC that Google’s AI deals with the U.S. and Israeli militaries crossed lines the company’s own ethics rules were supposed to hold.
The resignations are the most public rupture yet in a year that has steadily pushed Google’s AI lab from research into the defense business, and they put a human face on a question the whole industry keeps dodging, from OpenAI’s own AI catching an alleged spy in its boardroom to government purges of foreign models
Who left, and what they said
Liguori, who spent three years at DeepMind working on AI systems for routing electricity through power grids, told the BBC she felt “ashamed to work at Google” as the company’s military contracts grew. She said she had actually stopped working months earlier, combining sick leave and holiday time after Google announced its classified military AI agreement with the Pentagon in April.
“I no longer believe that our technology will have a net positive impact on society,” she said, in an interview that also ran in The Intercept. She pointed to two specific moves: the December launch of GenAI.mil, the Pentagon’s generative AI contracting vehicle, and Google raising its cloud security authorization to IL6, the level required to handle classified workloads. “AI for targeting is very dangerous; it creates a system where it is easier to escalate the brutality of war,” she said.
Kirsch’s objection is narrower and arguably harder for Google to answer. He told the BBC that Google’s governance and oversight are “not sufficient or meaningful enough to deploy AI safely and responsibly in military and classified settings.” His sharpest claim: Google appears to have no right to control or veto how the Pentagon uses its AI once delivered, and only limited visibility into how it is being deployed.
The contracts at the center of it
The Pentagon agreement, signed in April, was one of several AI deals the U.S. military struck this year with commercial labs. OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and the startup Reflection signed comparable arrangements to deploy AI across military logistics, planning, and battlefield operations, even as OpenAI’s models move into Google Cloud’s enterprise catalog. The Google contract permits models to be used for “any lawful government purpose” while barring autonomous weapons and domestic mass surveillance without appropriate human oversight, a formulation Liguori argues cannot meaningfully constrain targeting systems.
The second grievance is Project Nimbus, the $1.2 billion cloud and AI contract Google shares with Amazon to supply Israel’s government and military, signed in 2024. Google said at the time that Nimbus did not cover sensitive, classified, or weapons-related work. Liguori told The Intercept that internal posts she wrote about the Gaza war were removed by company moderators, and that her concerns received procedural responses rather than substantive discussion.
Google’s response, through spokesperson Tom Hewett, is that the company works with public sector organizations worldwide and follows a consistent review process that complies with its acceptable-use policies and AI principles, and that employees can raise concerns through channels protected by a strict anti-retaliation policy.
A pattern, not a moment
What makes these resignations different from a typical ethics flare-up is the trail of exits behind them. DeepMind employees voted to unionize in May. In February, Liguori helped organize a protest outside the office under banners reading “No Gemini for Genocide.” In September, engineer Robert O’Callahan quit because his team was making AI run faster, another protest departure. Two current employees told the BBC anonymously that “a lot of people have left the company that feel strongly about this,” some publicly and some silently.
Both Liguori and Kirsch said their contracts restrict them from publicly criticizing the company, Liguori telling the BBC hers barred criticism effectively for life, and both chose to speak anyway. That is the detail that will linger. Companies can replace researchers; they cannot easily un-say what on-the-record departures reveal about internal confidence. The U.S. Department of Homeland Security ordered federal agencies to purge DeepSeek’s AI from their systems this summer over security fears, showing governments are drawing hard lines around which AI runs inside sensitive infrastructure. The DeepMind exits suggest the people who build that AI are starting to draw their own lines too.
The test for Google now is the one Kirsch laid out: not whether its review process exists, but whether it can actually constrain what happens after the contract is signed. Nothing in the company’s statement answers that.
