Google DeepMind Researcher Resigns After Quiet Pentagon Contract

Alex Turner, an AI safety researcher at Google DeepMind, resigned from the company after Google's leadership signed a classified contract with the Pentagon to use Gemini models without legally binding restrictions on autonomous weapons and mass surveillance. Turner had spent months trying to block the deal from within.

What Happened

Turner, who worked as a research scientist at Google DeepMind, spent months organizing internal opposition to the contract the Pentagon was proposing to major AI companies. The US Department of Defense was asking suppliers to accept an 'all lawful use' clause, granting consent to use their systems for any lawful military purpose, without hard exclusions for autonomous lethal weapons or mass surveillance.

According to Turner's own account, published on his blog, when Google quietly signed such a contract on April 28, 2026, he concluded that staying at the company was no longer compatible with his own principles. He resigned shortly after.

The Plan That Stalled With Lawyers

Before the contract was signed, Turner had drafted his own proposal: a 25-page document titled 'Framework for Military AI Oversight.' It included hard, legally binding provisions excluding lethal autonomous weapons operating without human control, a ban on untargeted AI profiling of people, the creation of an independent oversight body for defense contracts, and annual transparency reports. The document went directly to Hassabis, who forwarded it to the company's policy team, where, according to Turner, it went unanswered until the contract was signed.

When Google signed the deal, my conscience just said 'no' - Alex Turner, former Google DeepMind researcher

Petition to Jeff Dean

In parallel, Turner organized an internal petition to Google's chief scientist, Jeff Dean, calling on him to oppose the deal. About 250 DeepMind employees signed it. Dean later signed a legal opinion supporting Anthropic's legal challenge to similar contracts, but according to Turner, he did not use his position to block Google's decision.

Among the leading AI companies, only Anthropic refused to sign a contract on the unconditional terms the Pentagon demanded. OpenAI, xAI, and other major suppliers agreed to similar terms, putting pressure on Google not to fall behind its competitors.

A Union as the Response

The Pentagon contract also led to an unprecedented step at DeepMind's UK division. In April 2026, employees voted to form a union backed by the Communication Workers Union and Unite the Union, with 98 percent support among those voting, the first such case at a lab working on the world's most advanced AI models.

The union is demanding, among other things, an end to Google's military AI being used by US and Israeli forces, the restoration of the company's previously withdrawn commitment not to use AI in weapons, the creation of an independent ethics oversight body, and the right of individual researchers to refuse work on projects they consider morally unacceptable.

A Repeat of 2018, Without the Same Force

The situation echoes the dispute over Project Maven in 2018, when nearly 4,000 Google employees signed an open letter against a Pentagon contract to analyze drone footage, and several left the company in protest. At the time, the pressure led Google to abandon renewing the contract.

This time, as former Google employee Laura Nolan points out, employee leverage is weaker. The tech industry has been through waves of layoffs, some of the internal communication channels that allowed staff to organize after 2018 have been dismantled, and other major AI companies have already accepted similar Pentagon terms, meaning Google is not an isolated case.

For Polish companies using Gemini or other systems from major AI providers, the episode is a reminder that suppliers' ethical principles are sometimes declarations without binding contractual terms, and that decisions about military applications are made beyond any control by commercial customers. In February 2025, Google removed from its published principles an earlier commitment not to use AI in weapons, which some employees read as a sign of weakening internal oversight.