'You feel you are being watched': DeepMind staff quit over Pentagon AI

Sofia Liguori joined Google DeepMind in 2023 to work on something hard to object to: using AI to route power through electrical grids. This week she quit, and she wrote her colleagues a letter that ended with a request to "stay human."

"I no longer believe that our technology will have a net positive impact on society," Liguori wrote in that letter, according to The Intercept. She told the BBC she is ashamed of Google's growing military work. She is one of several DeepMind staff who have resigned or are preparing to leave over the company's contracts with the US military, according to the BBC's reporting. She is also speaking out despite a contract that limits what she can say about her former employer. She told the BBC her exit terms bar her from criticising Google for life.

The other named departure is Andreas Kirsch, an AI researcher who left earlier in October. He told the BBC that Google's governance and oversight are not adequate for deploying AI in classified military settings, and that the company appears to have limited visibility into how the Pentagon actually uses its models.

"Any lawful government purpose"

The contract at the centre of this was signed earlier this year. It lets the Pentagon use Google's AI models for "any lawful government purpose," including classified military work, according to Ynetnews' summary of the BBC investigation. It includes restrictions on domestic mass surveillance and on autonomous weapons, including target selection, without appropriate human oversight and control.

That last clause is where the argument lives. A restriction on weapons that pick targets "without appropriate human oversight" is a restriction whose meaning depends entirely on who decides what "appropriate" is. Kirsch's point, as reported, is that it is unclear whether anyone at Google is in a position to check.

The political backdrop is not subtle. Defence Secretary Pete Hegseth has said he wants the military to become an "AI-first warfighting force." Google, which famously walked away from the Pentagon's Project Maven drone program in 2018 after staff protests, is now in the room.

The view from inside

Liguori's account to The Intercept goes back further than one contract, and it is mostly about what it felt like to object. She traced her doubts back to Project Nimbus, the cloud computing deal Google and Amazon won in 2021 to serve the Israeli government, including defence-related agencies. She joined the employee group No Tech for Apartheid in 2024. In February 2026 she helped organise a protest outside DeepMind's London office, with banners that read "No AI in the kill chain" and "No Gemini for Genocide." DeepMind staff voted to unionise in May.

She told The Intercept that some of her internal posts were deleted for containing the word "genocide," and that she was told the word was intrinsically violent and her tone inappropriate. HR tickets, she said, got automated replies. The day after Google announced its classified military contract in April, she stopped working, combining sick leave and holiday.

"You feel you are being watched, you can't speak," she told the BBC, as quoted by Ynetnews.

Her own view is that petitions were never going to work and that staff should have gone on strike. She also told The Intercept that fear and self-censorship are widespread at the company, and that the boundaries workers believe exist are narrower than they look.

Google's answer is procedural. Spokesperson Tom Hewett told The Intercept that the company "follow[s] a consistent process for reviewing and entering these contracts," and that it has "dedicated channels for employees to safely raise any concerns." Google says it has policies against retaliation, and that its contracts comply with its AI principles, including human supervision and respect for international law.

Both things can be true at once. There can be a process, and the people inside it can feel that the process exists mainly to absorb complaints. What stands out is who is leaving. These are the researchers DeepMind recruited partly by selling itself as the lab that cared most about doing AI carefully. Liguori came to optimise power grids. She is leaving because she thinks the grid she was helping to build now runs to a battlefield.

Nobody has published a count of how many have gone. Google has not said. For now the number that matters is the one in Liguori's contract: how long she is supposed to stay quiet. Her answer, apparently, is not very long at all.

Sources: BBC News, The Intercept, Ynetnews.