Google Deploys Gemini‑Powered AI Agents to Automate Pentagon’s Unclassified Tasks

Google’s rollout will begin with eight pre‑built agents that can digest meeting transcripts, draft budget spreadsheets and cross‑check proposed actions against the National Defense Strategy, according to Bloomberg. The tools run on the Pentagon’s unclassified networks, but Under Secretary of Defense for Research and Engineering Emil Michael told reporters that “talks are underway” to extend the same capability to classified and top‑secret systems (Bloomberg). By letting staff issue plain‑language commands, the agents aim to cut the time spent on routine paperwork that currently consumes a sizable fraction of the Department’s 3 million civilian and military workforce.

The deployment builds on a pilot that launched in December, when the Defense Department’s GenAI.mil portal gave 1.2 million users access to Google’s Gemini chatbot. Engadget reported that those users generated roughly 40 million unique prompts in the first six months, a volume that convinced senior officials the technology was ready for broader use. Google Vice President Jim Kelly said in a blog post that the new agents are “customizable via natural language,” allowing analysts and logisticians to create bespoke workflows without writing code (Engadget).

Beyond the pre‑packaged agents, the Pentagon will be able to train its own assistants using Gemini’s underlying large‑language model. The agency’s internal AI teams can feed domain‑specific data into the system, producing agents that understand defense‑specific terminology and processes. This approach mirrors Google’s recent “Auto Browse” feature in Chrome, which lets the Gemini engine navigate the web autonomously to retrieve information (Wired). While the Chrome tool is consumer‑focused, the underlying capability demonstrates the model’s ability to reason across multiple steps—a prerequisite for the multi‑task automation the Pentagon seeks.

Security considerations remain a focal point. The current rollout is confined to unclassified environments, and Google has pledged to harden the agents against data leakage and adversarial prompts before any classified deployment. According to the Engadget piece, the Department is already evaluating “robust access controls and audit trails” to ensure that sensitive information does not flow outside approved channels. Analysts familiar with defense procurement note that such safeguards are essential for any AI integration, given the heightened risk profile of military networks.

If the program scales as envisioned, the Pentagon could see a measurable reduction in administrative overhead, freeing personnel to focus on analysis and decision‑making. The move also signals a deeper partnership between the U.S. government and big‑tech AI firms, echoing Google’s broader push to embed Gemini across enterprise workflows—from Android task automation (TechCrunch) to virtual‑world reasoning (TechCrunch). As the first major U.S. defense client to adopt Gemini‑powered agents, the Pentagon’s experience will likely shape future contracts and set a benchmark for how generative AI can be safely leveraged in high‑stakes, government settings.