Google Signs Classified Gemini Deal with Pentagon

The Information reports that Google signed an amendment to its existing agreement with the US Department of Defense allowing Pentagon personnel to deploy the Gemini family of models for "any lawful government purpose," including on classified systems. Reuters and Yahoo Finance cite The Information in reporting the same contract language. Multiple outlets, including 9to5Google and The Verge, report that the agreement requires Google to "assist in adjusting its AI safety settings and filters at the government's request," language first reported by The Information. Reporting by The Information and Reuters states the contract also says Google does not have "any right to control or veto lawful government operational decision-making." The Washington Post and The Information report that more than 600 Google employees signed a letter asking CEO Sundar Pichai to bar classified work.

Editorial analysis - technical context: Deploying frontier models on classified networks typically involves separate operational environments, hardened infrastructure, and controls for data exfiltration and model access. For practitioners, this often means air-gapped or heavily segmented cloud enclaves, strict key and API governance, and additional logging and auditing layers. Adjusting model "safety settings and filters" in such settings can require offline configuration, custom prompt policies, or distillation and fine-tuning performed within the secured environment rather than via public APIs.

reporting places this agreement in a broader pattern of the US defense establishment securing access to commercial frontier models; Reuters noted the Pentagon signed similar arrangements with other labs in 2025. The Verge and Washington Post coverage highlight internal employee opposition at Google, echoing previous controversies such as the public debates around Anthropic and other providers. For policy watchers, the combination of contract language that disavows veto power and employee pushback tightens the debate over corporate governance, contractor obligations, and the scope of permissible government uses.

For practitioners: 1) whether the Department of Defense publishes or clarifies allowable use-cases and technical integration requirements for classified deployments; 2) how vendors implement and audit the "safety settings and filters" the contract references, especially in isolated environments; 3) whether employee petitions, Congressional scrutiny, or procurement rules lead to public changes in contract language or oversight mechanisms; 4) how other large AI vendors negotiate analogous terms and whether those deals include veto, auditing, or contractual restrictions on operational uses.

Reported facts in this note are drawn from The Information (as cited by 9to5Google and The Verge), Reuters, 9to5Google, The Verge, Yahoo Finance (Reuters reprint), and The Washington Post. Google provided a public statement quoted in reporting by 9to5Google and The Verge that begins "We are proud to be part of a broad consortium of leading AI labs and technology and cloud companies providing AI services and infrastructure in support of national security."