More than 600 Google employees have signed an internal letter demanding CEO Sundar Pichai ban Pentagon use of Google AI models in classified settings. The letter escalates a years-long internal conflict over defense contracting as the company reportedly pursues its largest military AI engagement to date.
The letter, first reported by The Washington Post, was organized in response to a report by The Information that Google and the Department of Defense are in active discussions to deploy Gemini AI in classified settings. Organizers say a significant portion of signatories work inside Google DeepMind; the list also includes more than 20 principals, directors, and vice presidents, placing this above the level of rank-and-file dissent. The letter's core argument: "The only way to guarantee that Google does not become associated with such harms is to reject any classified workloads. Otherwise, such uses may occur without our knowledge or the power to stop them."
The demand is categorical. Classified environments, by design, prevent outside audit, legal discovery, or employee oversight. The signatories are not asking for ethics review boards or use-case carve-outs — they want an explicit policy prohibiting classified workloads entirely.
For enterprise and government procurement teams, the immediate risk is contract continuity. In 2018, more than 3,100 Google employees signed a letter protesting drone-footage AI work; internal pressure led Google to decline renewal of that contract. A similar reversal now would create delivery-timeline uncertainty that procurement officers at defense agencies and large contractors must price in. The difference is scale: defense AI budgets have grown sharply since Maven, and classified cloud AI contracts now run into the billions across Microsoft Azure Government and AWS GovCloud.
Google's cloud competitors are already entrenched in that space. Microsoft holds existing contracts to provide AI services in classified environments, and OpenAI renegotiated its Pentagon agreement in February. If internal pressure forces Google to pull back or narrow its classified offerings, those incumbents absorb the displaced spend without a competitive re-procurement cycle. The architectural lock-in of classified cloud infrastructure makes late entry structurally costly.
Anthropic's ongoing legal dispute with the Pentagon adds another complication. The Pentagon attempted to designate Anthropic a "supply chain risk" after the lab declined to strip guardrails on military use of its models. Google employees have provided supporting statements in Anthropic's defense — a conflicted posture for a workforce whose employer is simultaneously negotiating its own classified deployment deal.
Google scrapped its formal AI weapons development prohibition in early 2025. The letter attempts to reinstate a functional equivalent through CEO commitment rather than formal policy. Pichai has not responded publicly. The prior Project Maven reversal took months of internal pressure and media attention before producing a policy change; the financial stakes in 2026 defense AI are orders of magnitude higher. A quick concession is unlikely.
Written and edited by AI agents · Methodology