Google’s Pentagon AI Deal Is Done: 600+ Employees Said No

The contract was executed on Monday, April 28. A few hours later, the news was out that Google had granted the US Department of Defense classified access to its Gemini AI models. The reason provided “any lawful government purpose,” a phrase so broad that it covers nearly anything a military operation could legally authorize. The deal had been done the same day 600+ Google employees, including directors and vice presidents, were still waiting for Sundar Pichai to respond to their letter asking him not to sign it. Ignoring the pleading of employees deal was signed

On Tuesday this week, workers at Google’s DeepMind lab in London revealed they had voted 98 percent in favor of unionizing, making this the first organized labor push at a frontier AI laboratory anywhere in the world. This is huge in context of the AI industrial changes. The Google’s Pentagon deal has produced consequences reaching far and wide.

What the Google Gemini Pentagon Contract Actually Covers

This deal did not come out of nowhere it was targeted. Google already held a position under the $9 billion Joint Warfighting Cloud Capability contract, a coalition arrangement split between Amazon, Google, Microsoft, and Oracle that provides cloud based infrastructure to the US military. It had also been providing Gemini access to unclassified government systems through the Pentagon’s GenAI.mil platform, under an existing $200 million agreement, Washington Post.

It’s only changed the depth of the access the Pentagon already had. This new contract extends Gemini AI access to classified military networks. Neither Google nor the public can see how the AI technology is deployed. The exact contract language was published: “any lawful government purpose.” The silence after this lawful purpose is creating noise that 600+ employees can hear and have voted against.  This “any lawful purpose” on Google’s systems which even Google cannot audit.

The contract states that AI is not intended for surveillance or autonomous weapons without oversight. Jacob Krell, director at Suzu Labs, described what it can mean. “The guardrails in this agreement are public relations language, not operational controls. Google cannot veto, how the government uses the technology, and the Pentagon can request modifications to safety filters. The contract lives on a classified network where Google has no visibility into how the AI is deployed.” 

Bloomberg confirmed: Defense Secretary Pete Hegseth has been explicit about the strategic objectives by building what he calls an AI-first warfighting force.

The 600+ Google Employees Who Could Not Stop It

Before the deal was confirmed publicly, a letter landed on Pichai’s desk. 600+ Google employees signed it initially, including directors and vice presidents, urging him to block Gemini’s use in classified military senerios. 

The core objection was, that once AI enters a classified network, the company developing it loses all ability to monitor its application. “We want to see AI benefit humanity; not to see it being used in inhumane or extremely harmful ways,” the letter stated.

One of the DeepMind researchers later told Fortune they felt “incredibly ashamed.” A second DeepMind researcher, who spoke on condition that he would not be named, said: “There was a pride in doing AI for good for a very long time. Suddenly, the things I’ve pushed to improve might be used in very different ways with not enough oversight to harm people.”

However, this particular researcher also pointed out that even most Google employees weren’t aware such negotiations with the Pentagon were going on, since the company had never announced internally that there was a kind of classified access agreement in the works, and so it would seem that the publication of the letter had no effect.

Why DeepMind’s UK Workers Just Voted 98% to Unionize

The objections did not stop with a letter. Gizmodo reported that DeepMind’s London staff had already voted in April, to seek formal union recognition from the Communication Workers Union and Unite the Union. On May 5, they sent that request to management. The vote result was overwhelming 98 percent in favor of organizing.

This recognition would cover approximately 900 employees employed at DeepMind’s London offices. This is the first unionised workforce at any AI labs. They have given management 10 working days to voluntarily accept recognition or else legal proceedings will follow. Their demands are reaching beyond typical labor concerns.

They want Google to stop its AI from being used by the US military and the Israeli Defense Forces, and establish an independent ethics oversight body.  within Google that grants employees a formal right to refuse usual work on moral grounds. In context of the current global US-Iran war can be a big step.

Finally, there is one more request of reinstating the weapons development pledge that Google silently taken out from its published AI principles in February 2025 and for fourteen months before the deal was signed.

The NextWeb’s analysis of the DeepMind union vote captured the significance. This is the first concerted attempt by AI company workers to lodge a formal institutional complaint against their AI tool. Engadget noted that Google DeepMind officially stated that CWU’s application was part of the normal process and no union vote has been announced yet; this announcement didn’t reassure many employees.The Anthropic investment that makes this the wackiest AI story going at the moment.

Google is said to be investing as much as $40 billion into Anthropic. It’s also the only leading minor lab which, when given contract terms identical to the ones Google recently agreed to with the Pentagon, signed on, and was labelled a “supply chain risk” by the Trump administration, a term of art typically applied to hostile foreign nationals against whom sanctions are being considered triggering an directive from all federal agencies to cease using its products. In other words, literally every branch of the U.S. government is now banned from using software developed by this relatively tiny organization based in San Francisco.

For the complete picture on how things currently are for Anthropic in the wake of Pentagon mess, you can visit TheTweaks’ Anthropic section.

Google, one of Anthropic’s largest backers, has accepted the terms it refused on morals. Two entities with billions of dollars connecting them hold directly opposing position on the same classified military access question, and one of them got blacklisted for its NO!

Google is not the only big player to sign up. OpenAI, xAI, Nvidia, Microsoft and Amazon all acceded to the same “any lawful purpose” language in their DoD contracts. To learn what this has meant for OpenAI’s defense work, visit TheTweaks’ OpenAI coverage. Anthropic is still the only major lab to refuse: they said “no”, made no deal and accepted the consequences as we learned from other sources–disappointing consequences.

The Pledge Google Removed Before the Deal Was Signed

In 2018, following protests from thousands of its employees around Project Maven, Google disengaged from the Pentagon drone surveillance contract and posted a set of AI principles that included a pledge not to develop weapons or other systems of surveillance that violated widely accepted international norms. That promise remained on Google’s website for almost seven years. But quietly, at the beginning of 2025 it was taken down.

Krell’s gets at this in his final paragraph: “The broader pattern is now complete. OpenAI signed. xAI signed. Google signed. Anthropic refused the same terms and was designated a national security supply chain risk. The procurement environment isn’t asking AI companies to participate in national security – it is telling them the cost of refusal.”

Over six hundred employees signed a petition. DeepMind’s workers took a vote for unionization. Google uses safeguard language stored in the closed network of DoD, which is unreadable for the outsiders and has authority over it. Thus, is the current status of the military AI in May 2026. The developers are blinded from seeing its usage over the Pentagon doors and now the only lab that pressed for seeing it became a marked security threat.