Google DeepMind resignations over military AI: who quit, the Pentagon and Nimbus contracts, and what Google says

Google DeepMind resignations over military AI became public on October 21, 2026, when the BBC and The Intercept published interviews with researchers who had quit over Google's work for the US and Israeli militaries. The BBC named two: research engineer Sofia Liguori, who left in the week of October 19, and AI researcher Andreas Kirsch, who resigned earlier in October. Their main objection is a 2026 agreement that lets the Pentagon use Gemini on classified networks for "any lawful government purpose." The BBC said "several" staff had resigned or were leaving. No total has been published.

This story is based on the BBC report as syndicated by Yahoo News, The Intercept's interview, Liguori's published resignation letter, Google's February 2025 blog post, and reporting by NBC News, 9to5Google, Axios, CNN and the Times of Israel, checked October 23, 2026. The BBC's own page couldn't be read for this story. The same fault line split Anthropic from the Pentagon earlier in 2026, as described in our Anthropic explainer.

The short version

• Who: Sofia Liguori, a London-based research engineer, after three years at Google; Andreas Kirsch, an AI researcher. Both spoke to the BBC.
• Why: Google's classified Pentagon deal for Gemini, signed in spring 2026, and Project Nimbus, its cloud contract with Israel's government.
• Speaking out: Liguori told the BBC her contract restricts her from criticizing Google, and that staff fear retaliation.
• Google's answer: its contracts are reviewed for compliance with its acceptable use policies and AI principles, and it has "strict workplace policies in place to prevent retaliation of any kind."
• Background: Google removed its pledge not to pursue AI weapons from its AI principles on February 4, 2025.

Who resigned from Google DeepMind, and what did they say?

Sofia Liguori joined in 2023 and first worked on AI for routing power through electrical grids, according to The Intercept. Her resignation letter, published on her own site, gives October 20, 2026, as her last day. "I no longer believe that our technology will have a net positive impact on society," she wrote, and "we should have said a much louder NO to our tech being used in the kill chain." She told the BBC the military work "makes me feel ashamed to work at Google." She told The Intercept that several of her internal posts were deleted because they used the word "genocide," and that "I was told that genocide is an intrinsically violent word."

Andreas Kirsch told the BBC that "Google's governance and oversight are not sufficient or meaningful enough," that "Google seems to have no right to control or veto how the Pentagon uses its AI," and that "current AI models are not robust enough" for such use.

They aren't the first. In June 2026, Business Today, citing Business Insider, reported that René Mayrhofer, Google's director of Android platform security, had resigned over the classified Pentagon deal. At OpenAI, robotics lead Caitlin Kalinowski quit in March over that company's Pentagon agreement, covered in our OpenAI humanoid story.

What contracts are they protesting?

Google's military AI work and the protests against it

1. Apr. 21, 2021Google and AWS win Israel's Project Nimbus cloud tender, about $1.2 billionTimes of IsraelConfirmed
2. Feb. 4, 2025Google rewrites its AI principles; weapons and surveillance pledge is goneGoogle; CNNConfirmed
3. Dec. 9, 2025Pentagon launches GenAI.mil with Gemini for Government for unclassified workAxiosConfirmed
4. Apr. 28, 2026Classified Pentagon deal for Gemini, 'any lawful government purpose'The Information via 9to5Google; NBC NewsReported
5. June 2026Android security director René Mayrhofer resigns over the dealBusiness Insider via Business TodayReported
6. Oct. 2026Andreas Kirsch, then Sofia Liguori, resign from Google DeepMindBBC; The Intercept; Liguori's letterConfirmed

• Confirmed
• Reported

The classified Pentagon deal. On April 28, 2026, 9to5Google relayed a report by The Information that Google had signed an amendment letting Pentagon staff use Gemini for classified work, for "any lawful government purpose." The Information said Google must "assist in adjusting its AI safety settings and filters at the government's request" and has no right to veto lawful operational decisions. On April 29, NBC News reported that a US official confirmed the agreement, and that Bloomberg had reported a letter from about 600 Google employees urging CEO Sundar Pichai to refuse. According to the BBC, the contract says the models shouldn't be used for domestic mass surveillance or autonomous weapons, including target selection, "without appropriate human oversight and control." Google hasn't published it.

GenAI.mil. On December 9, 2025, Axios reported, the Pentagon launched GenAI.mil with Google's Gemini for Government as its first major tool, for unclassified work. Liguori's letter says these steps made the classified deal foreseeable.

Project Nimbus. Google and Amazon Web Services won Israel's Nimbus cloud tender, valued at about $1.2 billion, in April 2021, the Times of Israel reported. When staff criticized the contract in 2024, Google said Nimbus "is not directed at highly sensitive, classified, or military workloads relevant to weapons or intelligence services," the BBC noted.

What does Google say?

A Google spokesperson gave the BBC and The Intercept the same statement: "We work with public sector organisations worldwide and follow a consistent process for reviewing and entering these contracts, including compliance with our acceptable use policies and AI principles." It added that Google has "dedicated channels for employees to safely raise any concerns" and "strict workplace policies in place to prevent retaliation of any kind."

In April, spokesperson Kate Dreyer told NBC News that Google remains committed to the consensus that AI shouldn't be used for "domestic mass surveillance or autonomous weaponry without appropriate human oversight."

How did Google's AI principles change?

In 2018, after thousands of employees protested its role in the Pentagon's Project Maven, Google declined to renew that contract, NBC News noted. Its 2018 principles listed applications Google would not pursue, including weapons and surveillance beyond international norms.

On February 4, 2025, James Manyika and Demis Hassabis published new principles built on three tenets, "Bold Innovation," "Responsible Development and Deployment" and "Collaborative Progress, Together." The post said "democracies should lead in AI development" and that AI should support "national security." CNN reported the same day that the list of applications Google wouldn't pursue was gone. Google's statements in 2026 still cite those principles. Emil Michael, the Pentagon official writing breach-reporting rules for AI labs as of October 2026, is covered in our story on the White House's Super Intelligence Force.

What's confirmed, reported and uncertain

• Confirmed: Liguori's resignation, last day and quotes (her letter, the BBC, The Intercept); Kirsch's resignation and quotes (BBC); Google's statement; the February 2025 rewrite (Google's blog, CNN); GenAI.mil (Axios); the 2021 Nimbus award (Times of Israel).
• Reported: the classified deal's terms (The Information via 9to5Google; the BBC; NBC News citing an official); the 600-employee letter (Bloomberg via NBC News); Mayrhofer's resignation (Business Insider via Business Today); Liguori's account of deleted posts.
• Uncertain: how many people have resigned; the full contract text; what the employment restrictions Liguori describes actually say.