Google DeepMind Resignations: Why Researchers Are Quitting Over Pentagon AI Work

Google DeepMind is losing researchers over military AI. Research engineer Sofia Liguori resigned this week, AI researcher Andreas Kirsch left earlier in October, and safety researcher Alex Turner quit in July after Google signed a Pentagon deal for classified AI work. The departing staff say Google has too little oversight of how the military uses its models. Google says its public-sector contracts go through a consistent compliance review.

Who has resigned from Google DeepMind, and why?

Three researchers have publicly resigned over Google's military AI work:

1. Sofia Liguori, a London-based research engineer who joined in 2023 and worked on AI for routing power through electrical grids, resigned this week. She told the BBC that pushing AI into military operations "makes me ashamed to work for Google," according to Il Sole 24 Ore's summary of the BBC report. In an internal letter quoted by The Intercept, she wrote that she no longer believed DeepMind's technology would have a net positive impact on society.
2. Andreas Kirsch, an AI researcher, resigned earlier in October. The BBC reports he said Google "seems to have no right to control or veto how the Pentagon uses its AI." He also argued that current models are not robust enough for high-stakes military use.
3. Alex Turner, who spent more than two years on AI safety at DeepMind, announced in a July 15 blog post that he was leaving. He wrote, "I couldn't stay at Google in good conscience, so I left," according to the Free Press Journal.

Outlets differ on Turner's title (research scientist or AI safety researcher), and none we reviewed gives Kirsch's exact title.

What does the Pentagon deal actually allow?

The deal lets Google's AI be used in classified military settings for any lawful government purpose, with some stated limits. The BBC reports the agreement says the models should not be used for domestic mass surveillance or for autonomous weapons, including target selection, without appropriate human oversight and control. The BBC says OpenAI, Amazon, Microsoft and others signed similar deals.

Coverage describes the restrictions in different ways:

• The OECD.AI incident record for Turner's resignation calls the deal "unrestricted." Citing press reports, it says the contract places no restrictions on autonomous weapons or surveillance. It classifies the case as an AI hazard because no specific harm has been reported.
• The Free Press Journal, summarizing Turner's account, says the contract states the system is not intended for those uses but gives Google no right to veto lawful operational decisions. Turner argues those terms are too weak.
• Reports also differ on timing. The Intercept says Google announced the classified contract in April. The Free Press Journal says the Pentagon confirmed its agreements with several companies in early May.

How does Google respond to the resignations?

Google says its public-sector work follows a consistent review process. A spokesperson told the BBC that contracts are reviewed for compliance with its acceptable use policies and AI Principles, and that employees have dedicated channels for raising concerns, with policies against retaliation.

Google and its supporters have also made a broader case for defense work:

• Democratic leadership in AI. When Google revised its AI principles in February 2025, James Manyika and Demis Hassabis wrote, "We believe democracies should lead in AI development," as reported by LBC.
• Human oversight built in. The revised principles commit to appropriate human oversight, due diligence and feedback mechanisms, and the contract wording reported by the BBC repeats that standard.
• Speed as a security need. Pentagon officials argue that rapid adoption matters. The BBC reports Under Secretary Emil Michael argued that speed decides who wins in the AI era.

Critics quoted by the BBC, including retired Lt. Gen. Jack Shanahan, former director of Project Maven, question whether people can meaningfully check AI recommendations under time pressure and whether deployment is outpacing testing.

What background explains why this keeps happening at Google?

Google has a long history of employee pushback on military work, and its own rules have changed over time:

• 2018, Project Maven. After thousands of employees signed a petition and dozens reportedly quit, Google said it would not renew its Pentagon drone-imagery contract, Engadget reported. Google published its first AI principles that year.
• 2021 onward, Project Nimbus. Google and Amazon hold a roughly $1.2 billion cloud and AI contract with Israel's government. Google has said Nimbus is not aimed at highly sensitive, classified or military workloads. The BBC copy we reviewed calls it a 2024 contract, while other reporting dates it to 2021.
• February 2025, AI principles rewrite. Google removed its list of "applications we will not pursue," which had included weapons, TechCrunch reported.
• 2026, organizing. The Intercept reports that DeepMind staff voted to unionize in May. The Free Press Journal reports that about 600 Google employees signed an April petition against classified work.

Why does an employer-brand controversy matter for how a company is described?

These stories now shape how a company is described in search results and AI-generated answers, not just in the news. When someone asks an AI assistant about working at Google DeepMind, the answer draws on the sources cited here: BBC investigations, resignation letters and OECD records. A few vivid quotes can become the default description for months.

This holds for any brand, not only big tech:

• Narratives harden fast. A resignation letter that gets widely syndicated becomes a source that AI models cite again and again.
• Fair context often gets lost. Google's reasoning may appear less often in AI answers than critics' quotes.
• Employer brand feeds buyer perception. Candidates, customers and partners increasingly ask the same assistants.

Teams can track this systematically. Our guide to auditing how AI describes your brand covers the method. Model Share measures how often and how favorably assistants mention you, and the GEO guide explains how to publish clear, accurate source material that AI models can pick up. For related safety coverage, see our report on OpenAI agents and disabled misalignment monitors.

FAQ

Who is Sofia Liguori?

Sofia Liguori is a research engineer who spent about three years at Google DeepMind in London. She resigned in October 2026, citing Google's military contracts with the US and Israel.

Why did Andreas Kirsch leave Google DeepMind?

According to the BBC, Kirsch said Google's oversight is inadequate for classified military use and that current models are not robust enough for high-stakes settings.

Why did Alex Turner quit Google DeepMind?

Turner left in July 2026 after Google signed its Pentagon deal. He says management had insisted Google would not sign, and his proposed safeguards went unaddressed.

Did Google drop its pledge not to use AI for weapons?

Yes. In February 2025, Google removed the section of its AI principles that listed applications it would not pursue, including weapons and certain surveillance. It replaced that section with commitments to human oversight and due diligence.

What is Google's position on military AI contracts?

Google says it reviews public-sector contracts for compliance with its policies and AI Principles. Its leaders argue that democracies should lead in AI development.

How can brands track how AI assistants describe them during a controversy?

Regularly ask major assistants the questions your audience asks, record the sources and tone, and publish accurate information that fills the gaps.

Last updated October 23, 2026.