Google will provide the Pentagon with AI agents built on Gemini

Google launches AI agents for the U.S. Department of Defense

* Who and why?

Google is beginning to use agents based on Gemini—a large language model—to serve more than 3 million DoD employees (military and civilian).

* First steps

In the initial phase, agents operate in open networks. Negotiations are underway to expand access to secure infrastructures where classified documents are processed.

* Key figures

Deputy Secretary of Defense for Research and Engineering Emil Michael announced the launch. Google Vice President Jim Kelly noted that the agency will be able to create its own agents by giving them tasks in natural language.

* Agent functions

Four pre‑built agents are deployed for:

- summarizing meeting notes;
- preparing budgets;
- reviewing action proposals for compliance with national security strategy.

* Pentagon portal

Since December of last year, the Google chatbot hosted on the portal has served 1.2 million DoD employees in unsecured networks:

- 40 million unique queries;
- over 4 million documents uploaded.

* Training staff

As of December, only 26,000 people have been trained, even though demand for sessions is fully booked—indicating growing employee engagement with AI.

* Context of collaboration

The partnership expansion follows a conflict between the DoD and Anthropic. The latter refused to remove restrictions that prohibit using its models for surveillance or autonomous weapons. Consequently, the DoD added Anthropic to its list of supply‑chain risk organizations; the company plans to challenge this decision in court.

* Industry reaction

About 900 Google employees and 100 OpenAI employees signed an open letter calling for similar restrictions. In February, Google quietly revised its “AI Principles” to reflect these constraints.