Google lands Pentagon Gemini deal

Google’s latest Pentagon deal is about one thing: moving Gemini from office-style government work into classified military networks. That matters because “classified” changes both the technical setup and the oversight problem. Once a model sits inside secret systems, outsiders can’t really see what it is helping with. This week, Google crossed that line. ### What actually changed? Google and the Pentagon reached a new agreement that lets the Defense Department use Gemini on classified networks for “any lawful government purpose.” That phrase is the real headline. It is broad on purpose — not a narrow pilot, not a single research program, and not limited to back-office chores. NBC says the deal was finalized this week, after earlier reporting that Google had been negotiating an expansion of its defense work. (nbcnews.com) ### Was Google already working with the Pentagon? Yes — but in a more limited way. Google had already gotten Gemini for Government onto GenAI.mil, the Pentagon’s enterprise generative-AI platform, for unclassified use. Google’s own public-sector post said that rollout started in December 2025, and the Defense Department described Gemini as the first frontier model on the platform. So the new move is not “Google enters defense.” It is “Google moves deeper into defense.” (cloud.google.com) ### Why does “classified” matter so much? Because classified use is where the hardest questions start. A chatbot helping draft memos is one thing. A model available inside secret military systems is another. Even if nobody says “weapons targeting” out loud, broad contract language can cover intelligence analysis, planning, logistics, and other operational work. The catch is that public guardrails become harder to test when the workloads are secret by design. That is exactly why employees focused on the classified part, not just the Pentagon part. (9to5google.com) ### Why are Google employees upset? More than 600 Google employees signed a letter urging Sundar Pichai to reject the deal. The signers came from DeepMind, Cloud, and other parts of the company, and the letter warned that classified military AI work could cause “irreparable damage” to Google’s reputation and make harmful uses harder to detect. Some senior staff signed openly, which gives the protest more weight than a quiet internal grumble. (theinformation.com) ### Why is this fight happening now? Because Google changed its own posture first. In February 2025, the company removed earlier language in its AI principles that had explicitly ruled out weapons and certain surveillance uses. That did not automatically mean “build military AI,” but it cleared away a bright-line ban that employees had relied on since the Project Maven backlash years ago. The Pentagon deal looks, to critics inside Google, like the practical consequence of that policy shift. (cnbc.com) ### Is Google unusual here? Not really — and that is part of the story. Big AI companies increasingly want defense business, especially as governments look for frontier models that can run in secure environments. But vendors are splitting over how many restrictions they are willing to demand. Several reports frame Google’s win partly as an opening created after Pentagon friction with Anthropic over military-use limits. In plain English: the Defense Department wants flexibility, and Google was willing to give more of it. (msn.com) ### What does the Pentagon get from this? Speed, mostly. GenAI.mil is supposed to give a huge defense workforce access to generative AI tools without every unit building its own stack from scratch. Defense One said the platform’s usage is already climbing and that newer Gemini models were added in April. For the Pentagon, a classified Gemini deal means the same basic promise can now extend into more sensitive work. For Google, it means a deeper foothold in one of the most strategically important AI customers on earth. (defenseone.com) ### So what is the real stakes question? It is not whether Gemini will write a few military emails. It is who gets to define acceptable military AI use when the systems are powerful, the contracts are broad, and the workloads are secret. Google just made its answer clearer. The company wants to be inside the room. (nbcnews.com)