Google–Pentagon Gemini talks

Google is negotiating with the Pentagon over a deal that would let Gemini run inside classified military systems. (money.usnews.com) Reuters, citing The Information, reported on April 16 that the talks would permit the Department of Defense to deploy Gemini in classified settings and could cover “all lawful uses.” The report said Google has pushed for contract language barring domestic mass surveillance and autonomous weapons without appropriate human control. (money.usnews.com) The Pentagon did not confirm negotiations, but a Defense official told Reuters the department will keep deploying frontier AI through industry partnerships “across all classification levels.” Google did not immediately respond to Reuters’ request for comment. (money.usnews.com) Classified settings are the government’s locked rooms for secret work, where software must meet extra security rules before it can be used. Google said on May 28, 2025 that its Google Distributed Cloud and air-gapped appliance won Department of Defense Impact Level 6 authorization for Secret data, building on earlier Impact Level 5 and Top Secret accreditations. (cloud.google.com) Google said that same 2025 milestone made Vertex AI and Gemini available at Impact Level 6 and Top Secret, meaning the company already had accredited infrastructure for high-classification workloads before these talks surfaced. Its cloud unit also said those systems can run in Google data centers or in forward locations for defense and intelligence customers. (cloud.google.com) The Pentagon has been widening its use of generative AI beyond office tools and into military and intelligence work. The Chief Digital and Artificial Intelligence Office says its GenAI.mil program is putting AI models in the hands of civilian and military personnel “at all classification levels,” alongside programs for battle management, targeting support, and data fusion. (ai.mil) That push has already moved past simple question-answering. MIT Technology Review reported on March 17 that the Pentagon was discussing secure environments where companies could train military-specific AI models on classified data, and said models from Anthropic were already being used to answer questions in classified settings. (technologyreview.com) For Google, the talks reopen a relationship that blew up inside the company in 2018. More than 3,000 employees signed a letter demanding Google cancel Project Maven, a Pentagon program, and CNBC reported at the time that workers warned the work would tie the company to warfare technology. (cnbc.com) Google’s public guardrails have also shifted since then. CNBC reported on February 4, 2025 that Google removed language from its AI principles saying it would not pursue weapons or surveillance uses, replacing the older pledge with a broader responsible-AI framework. (cnbc.com) If the Gemini talks produce a deal, Google would be entering a Pentagon AI market that now spans unclassified chatbots, Secret cloud systems, and model work tied to battlefield decisions. If they do not, the reports still show how far the Pentagon’s search for classified AI partners has moved into Big Tech’s core products. (money.usnews.com)