Google Signs Classified Pentagon AI Deal Amid Employee Revolt

Google has signed an amendment to its existing Pentagon contract, giving the Department of Defense access to its Gemini AI models on classified military networks. The deal, first reported by The Information and confirmed by Bloomberg, NBC News, and TechCrunch, extends the company's existing $200 million DoD contract from unclassified systems into classified networks used for sensitive operations, including mission planning and weapons targeting.

The contract language authorizes use for "any lawful government purpose," with a written condition that the AI "is not intended for, and should not be used for, domestic mass surveillance or autonomous weapons (including target selection) without appropriate human oversight." In the same document, Google agreed it holds no right to "control or veto lawful government operational decision-making" and committed to adjusting the AI's safety filters on the government's request.

Inside Google, the reaction was not quiet. More than 700 employees signed a letter to CEO Sundar Pichai this week asking the company to reject the classified work, warning of "irreparable damage" to the company's reputation and values. It is the most significant internal opposition since 2018, when Project Maven protests led Google to exit a Pentagon contract for AI-assisted drone surveillance. This time, Google did not walk away.

Stepping in where Anthropic stepped back

The timing is not coincidental. Anthropic's Mythos model, built for high-stakes security applications, was effectively blocked from government use earlier this year after being flagged as a supply-chain risk. The Pentagon's AI chief confirmed this week that the department is now working with Google, OpenAI, and xAI, and was explicit about the reasoning: "reliance on one model is never a good thing."

Google reportedly also exited a separate $100 million drone swarm program around the same time, according to Tom's Hardware, though it is not clear whether that withdrawal was tied to the classified contract or was a separate negotiation.

The safety commitment question

What the contract text actually shows is that Google's safety commitments here are adjustable. The company can flag concerns. The government defines what is lawful. When those positions diverge, the contract does not specify what Google can do about it. The prohibition on autonomous weapons without human oversight is real on paper, but the definition of adequate human oversight is not pinned down.

That is the distinction the 700 employee signatories were raising: not that Google is building weapons outright, but that agreeing to adjust safety filters on request is a meaningfully different thing from maintaining an independent safety standard. Whether Google's leadership sees it the same way is not yet clear.