Google signs classified Pentagon deal

Alphabet’s Google has signed a classified agreement allowing the Pentagon to use its artificial intelligence models for sensitive government work. (reuters.com) Reuters reported on April 28 that the deal lets the Department of Defense use Google’s AI for “any lawful government purpose.” The report said that includes classified networks used for mission planning and weapons targeting. (reuters.com) The same Reuters report said the Pentagon signed agreements worth up to $200 million each with major artificial intelligence labs in 2025, including Google, OpenAI and Anthropic. Google told Reuters it supports government agencies on classified and unclassified projects. (reuters.com) Artificial intelligence models like Gemini work by predicting patterns in text, images and data, then generating answers, summaries or recommendations from those patterns. On classified systems, that can mean helping analysts sort information faster or draft options for military planners. (google.com) (reuters.com) The contract language described by Reuters says the system is not intended for domestic mass surveillance or autonomous weapons without human oversight and control. The same agreement also says Google does not get the right to control or veto lawful government operational decision-making. (reuters.com) CNBC reported the Pentagon has expanded its use of Google’s Gemini after dropping Anthropic about two months earlier and labeling it a supply chain risk. Pentagon artificial intelligence chief Cameron Stanley told CNBC that relying on one vendor is “never a good thing.” (cnbc.com) CNBC also reported that more than 700 Google employees signed a letter to Chief Executive Sundar Pichai this week urging the company to reject classified workloads. CNET separately reported that the employee pushback centered on military and surveillance risks. (cnbc.com) (cnet.com) The backdrop is Google’s own rule change. In February 2025, CNBC reported that Google removed a previous pledge not to use artificial intelligence for weapons or surveillance from its public AI principles. (cnbc.com) Google’s current AI principles say it will use human oversight, testing and safeguards while aligning with international law and human rights. The Pentagon has said it does not want AI for mass surveillance of Americans or weapons that operate without human involvement. (google.com) (reuters.com) The immediate result is that Google now sits alongside OpenAI and xAI as a supplier of AI for classified Pentagon use, while Anthropic remains outside Defense Department work during its court fight. The next test is whether the safeguards described in public reporting hold once the work moves behind classified walls. (reuters.com) (cnbc.com)