Google signs classified Pentagon AI

Google has signed a classified artificial intelligence deal with the Pentagon that lets the Defense Department use its models for “any lawful government purpose.” (theverge.com) (thenextweb.com) The agreement puts Google alongside OpenAI and xAI as commercial AI suppliers cleared for classified Pentagon work, according to reports citing people familiar with the matter. Classified networks are used for sensitive tasks including mission planning and weapons targeting. (theverge.com) (usnews.com) The deal surfaced as more than 600 Google employees sent Chief Executive Sundar Pichai an open letter asking him to reject classified military AI contracts. Organizers said signers came from Google DeepMind, Google Cloud and other units, including more than 20 directors, senior directors and vice presidents. (businessinsider.com) (euronews.com) In that letter, employees said opaque classified systems make outside oversight hard and warned Google’s tools could be used for mass surveillance or lethal autonomous weapons. One organizing employee said there was “no way to ensure” the tools would not be used to cause harm outside public scrutiny. (euronews.com) (businessinsider.com) Google’s reported contract language includes restrictions on domestic mass surveillance and autonomous weapons without appropriate human control, but reports say the Pentagon pushed for broader “all lawful uses” wording. Employees argued those limits would be difficult to enforce once models were running inside classified environments. (euronews.com) (thenextweb.com) The fight revives a dispute Google appeared to settle in 2018, when employee protests over Project Maven pushed the company to stop renewing a Pentagon contract tied to drone image analysis. That earlier revolt involved more than 3,000 employees and helped produce Google’s original AI rules. (cnbc.com) (pbs.org) Those rules changed in February 2025, when Google removed explicit promises not to use artificial intelligence for weapons or surveillance from its AI Principles page. Google’s current principles emphasize responsible development but no longer list those old prohibited categories. (cnbc.com) (ai.google) The Pentagon’s side of the dispute has also hardened. Euronews reported that officials argued broad lawful-use language was needed for operational flexibility, while earlier reporting said Anthropic lost Defense Department work after resisting unrestricted military uses of its systems. (euronews.com) (thehill.com) Google has not publicly released the full contract terms, so the most sensitive details still rest on reporting from outlets citing unnamed sources. But the direction is clear: the company that once walked away from Project Maven is now back inside classified Pentagon AI work. (theverge.com) (the-decoder.com)