Google DeepMind Researchers Quit Over US and Israeli Military AI Work

Google DeepMind staff are resigning over the company's classified Pentagon AI deal and its work with the US and Israeli militaries, and some spoke to the BBC despite contract limits.

By USA 24 Editorial TeamOctober 22, 20265 min read

Empty desk in a glass-walled office at dusk, a cardboard box of belongings beside a lit lamp, rain on the window. — Illustration: USA 24

ShareX Facebook LinkedIn Reddit Email

Key takeaways

Sofia Liguori, a London-based research engineer, resigned this week after three years at Google, saying the company's work with the US and Israeli militaries made her ashamed to work there.
Andreas Kirsch, an AI researcher, resigned earlier in October and told the BBC that Google's governance and oversight are 'not sufficient or meaningful enough' for military and classified AI.
The Pentagon contract reportedly allows use of Google models for 'any lawful government purpose' while explicitly barring autonomous weapons and mass surveillance without human oversight.
Both workers said non-disparagement clauses in their contracts stopped them from criticizing Google publicly, so they spoke to the BBC anyway.
Reports disagree on when the classified-network deal was signed, April or May, and on whether the $1.2 billion Israel-linked contract involves Google or Amazon.

Google DeepMind researchers are quitting over the company's military AI work, and the dispute has spread well beyond the lab. A BBC investigation published Wednesday, October 21 gathered testimony from current and former DeepMind staff. The same day, The Intercept published its own story on a resignation tied to a Pentagon contract.

The departures are not hypothetical. Sofia Liguori, a London-based research engineer, left this week after three years at Google. Ynet reported that she said the company's work with the US and Israeli militaries made her ashamed to work there. Reporting also says other staff left over weak safeguards and fears of retaliation, and that more are preparing to go.

The fight centers on a classified Pentagon AI deal and a set of older contracts that workers have objected to for years. It also raises a question that reaches beyond Google: can employees at AI labs criticize government work when their own contracts limit what they can say?

Who resigned and what they said

Liguori's objection goes further than doubts about the reliability of AI or autonomous weapons. Superpower Daily reports that she said internal petitions had limited effect. In a letter to colleagues, as reported by The Intercept, she wrote: "I no longer believe that our technology will have a net positive impact on society." In the same letter, she urged colleagues to "stay human" while the company pursued military work.

Andreas Kirsch, a former AI researcher at Google DeepMind, resigned earlier in October. He told the BBC that Google's governance and oversight are "not sufficient or meaningful enough" for AI used in military and classified settings. His argument is narrower than Liguori's moral objection, but it targets the same question: who decides how models get deployed, and with what checks.

Both workers said their contracts' non-disparagement clauses stopped them from criticizing Google publicly. They spoke to the BBC anyway. One current employee told the BBC there is a pipeline of potential departures, but the number is unknown.

What the Pentagon contract allows and what it bars

According to Kenya Today, the Pentagon contract reportedly lets the government use Google models for "any lawful government purpose." It explicitly bars autonomous weapons and mass surveillance without human oversight. Those limits read like guardrails, and they are the language Google points to when it defends the deal.

The workers' complaint, as reported, is that the safeguards are weak and that raising concerns carries a risk of retaliation. A contract can prohibit a use, but the company's internal process decides whether anyone checks compliance and what happens to the person who asks. That gap is the core of the dispute.

The Pentagon has pushed hard for AI adoption. Defense Secretary Pete Hegseth wants the US to become an "AI-first warfighting force," according to coverage collected by Kahawatungu, and he has made integrating AI into the armed forces a top priority. For Google, a classified contract is a major commercial opportunity, and it is also a recruiting problem.

The Israel-linked contract and the timing questions

Reports put a related cloud contract tied to Israel at $1.2 billion, but the sources disagree on who the counterparty is. Ynet describes a Google cloud agreement with Israel. Zetik calls it the Project Nimbus contract with Amazon. Until official documents or company statements settle the framing, readers should treat the counterparty as disputed.

Zetik also cites a 2024 letter from nearly 200 DeepMind workers that objected to Google's military contracts, including Israeli military cloud work. That makes this a recurring fight, not a one-week flare-up.

The timeline is also inconsistent. One account says the Pentagon classified-network deal was signed in April. Superpower Daily reports that the Pentagon named Google among eight companies agreeing to provide AI capabilities for classified networks in May. The exact sequence remains unresolved.

How Google responded

Google told the BBC that it follows a consistent contracting process, complies with its AI Principles, and offers channels for raising concerns without retaliation. A company spokesperson said Google "follow[s] a consistent process for reviewing and entering these contracts." That is a spokesperson statement, not a quote from an executive.

The response does not fully answer the allegations about internal dissent. The Intercept asked Google about Liguori's criticism, and an editor's note reportedly suggests the company's answer was incomplete. That matters because the workers' central charge is not only about contract terms. It is about how the company handles people who object.

Why this fight keeps coming back

The pattern is familiar. A lab builds general-purpose models, a government customer wants them for sensitive work, and employees who joined to build consumer or research products object. Military AI is also moving from experiments to programs. The Army's ECHOTRIBBLE proposal, which would deploy worm-like AI agents, shows how quickly defense planners are testing autonomous software.

Oversight of AI agents is becoming a dispute inside companies too. In a separate story, reporting described weakened safety monitoring at OpenAI, and one side denied it. The common thread is that outside observers often learn about internal safety debates only after someone leaves.

The near-term winners are easy to name. Google and the Pentagon gain a path to deploy frontier models in classified work. The losers are less clear. Workers who want to speak up face contract limits and a fear of retaliation, and Google risks talent losses and public trust if departures continue.

What tech workers can check now

The story's most practical lesson is for people who work at AI companies. A few steps are worth taking:

Read any non-disparagement or confidentiality clause before signing, and check whether it reaches disclosures to regulators or lawmakers.
Put internal concerns in writing, and note who received them and when.
Find your company's formal reporting channels and confirm they cover concerns about government contracts.
Talk to an employment lawyer before speaking publicly if a clause is unclear.

What to watch next

Three things will show whether this becomes a broader revolt or a contained episode. First, whether more resignations are named on the record, since the BBC source described a pipeline of departures without a count. Second, whether Google or the Pentagon publicly confirm the signing date and the scope of the classified deal. Third, whether the Israel-linked contract's counterparty is settled in official documents. If the contract terms stay opaque, the next resignation letter will likely carry the same argument this one did, and the company will face the same question about how it hears dissent.

Frequently asked questions

Who resigned from Google DeepMind over military AI work?

Sofia Liguori, a London-based research engineer, resigned this week after three years at Google. Andreas Kirsch, a former AI researcher at Google DeepMind, resigned earlier in October. Both spoke to the BBC even though their non-disparagement clauses limited public criticism.

What does Google's Pentagon AI contract allow?

Reports say the contract lets the Pentagon use Google models for "any lawful government purpose." It explicitly bars autonomous weapons and mass surveillance without human oversight. Staff who left say the safeguards are weak, and reports disagree on whether the deal was signed in April or May.

Is the $1.2 billion Israel-linked contract with Google or Amazon?

The sources disagree. Ynet describes a Google cloud agreement with Israel, while Zetik calls it the Project Nimbus contract with Amazon. Until official documents settle the counterparty, the framing should be treated as disputed.

How did Google respond to the resignations?

Google told the BBC that it follows a consistent contracting process, complies with its AI Principles, and offers channels for raising concerns without retaliation. The Intercept reported that Google's response to specific criticism from Liguori was contested, and the editor's note suggests the answer was incomplete.

Sources

#Google DeepMind #Military AI #AI Ethics #Pentagon Contracts #Tech Workers

About this article. Reported and produced by the USA 24 newsroom using the sources cited above. Read our editorial policy. Spotted an error? Tell us and we will correct it.