The Department of Defense is scaling its use of Google's Gemini model for classified projects, Pentagon AI chief Cameron Stanley confirmed to CNBC. The move comes roughly two months after the DOD dropped Anthropic, citing supply chain risk—a decision now tangled in federal litigation with split rulings across two courts.

Stanley emphasized that the DOD is not betting on a single vendor. "Overreliance on one model is never a good thing," he said, noting the department is also working with OpenAI and other providers to modernize wartime capabilities. The architecture is deliberately heterogeneous: different models for different threat surfaces, latency requirements, and classification levels.

Internally, Google faces pushback. Over 700 employees signed a letter to CEO Sundar Pichai this week, opposing classified workloads on ethical grounds. For the DOD, the calculus is operational efficiency. Stanley claims Gemini is saving "thousands of man hours weekly" across the force.

Anthropic's exclusion persists despite a preliminary injunction from a San Francisco judge barring enforcement of the ban at other agencies. The DOD spokesperson confirmed Anthropic is not currently under contract, though President Trump hinted a deal might be possible.

Stanley framed the recent limited release of Anthropic's Mythos model as a wake-up call, citing its advanced cyber capabilities and associated risks. "You don't cook a Thanksgiving turkey in the microwave," he said, advocating for matching technology to use case. The DOD is now building pipeline infrastructure to support multiple LLMs, each serving distinct mission profiles—a design choice that trades vendor lock-in for integration complexity.

Source: CNBC