Google Rolls Out Gemini AI Agents to Pentagon's 3M Workers

Google is rolling out Gemini-powered AI agents across the Pentagon's three-million-strong workforce, moving beyond the chatbot phase into autonomous task execution. The Department of Defense on Tuesday unveiled Agent Designer, a no-code tool on its GenAI.mil platform that lets employees build custom AI assistants to automate multi-step workflows - no programming background required.

What Agent Designer Does

Agent Designer ships with eight pre-built agents that automate routine Pentagon work: summarizing meeting notes, building budgets, checking proposed actions against the national defense strategy, and generating planning and resourcing estimates for military tasks. Beyond the pre-built options, any DoD employee can create custom agents using natural language prompts.

The custom agents can ingest multiple data sources, execute multi-step automated tasks, and be shared across teams for immediate deployment. The Pentagon described specific use cases: operators generating after-action reports, analysts synthesizing Controlled Unclassified Information imagery into memos, and comptrollers building financial analysis applications.

"It saves you a lot of time in the middle, but you have to review at the end to make sure there's no hallucinations," said Emil Michael, Under Secretary of Defense for Research and Engineering.

Scale and Adoption

GenAI.mil has seen rapid uptake since its December launch under a contract brokered between the DoD and Google Cloud. The numbers so far:

The training gap stands out. While over a million employees jumped on the chatbot, only 26,000 have completed formal AI training - despite future sessions being fully booked. Michael emphasized training is "critical for preventing errors," a concern that grows sharper as the platform moves from a chatbot to autonomous agents.

Fort Bragg's Mission Training Complex offered one proof point for the efficiency gains: military exercise scenarios that previously required nine team members and six months were completed in six weeks using the AI portal.

Classified Expansion

The agents currently operate only on unclassified networks. Michael indicated that expansion to classified and top-secret systems is next: "We're starting with unclassified because that's where most of the users are, and then we'll get to classified and top secret."

Talks with Google over classified cloud deployment are underway. The DoD also plans to integrate tools from OpenAI and xAI into the GenAI.mil platform, though no timeline has been set for those additions.

The Competitive Context

The timing is hard to ignore. Google's agent rollout arrives days after the Pentagon designated Anthropic a supply chain risk for refusing to remove AI guardrails on autonomous weapons and mass surveillance. Anthropic was previously the only frontier AI lab cleared for use on classified Pentagon networks. With Anthropic suing the government and its classified access revoked, Google is filling the vacuum - and on far more favorable terms.

OpenAI and xAI have also secured Pentagon deals in recent weeks, all of them without the kind of use restrictions that cost Anthropic its government business.

Internal Dissent

Google's Pentagon push is not without friction inside the company. More than 200 Google employees - alongside nearly 50 OpenAI employees - signed a public letter in late February criticizing the Pentagon's negotiating approach with AI companies and urging their employers to avoid repeating Anthropic's treatment.

A separate internal letter, sent to Google Chief Scientist Jeff Dean on March 1, specifically asked the company to prohibit the military from using Gemini for surveillance of American citizens or for operating autonomous weapons without human oversight.

Dean publicly backed the concerns. "Mass surveillance violates the Fourth Amendment and has a chilling effect on freedom of expression," he wrote on social media - a notable statement from Google's most senior AI scientist, given that his employer just expanded its military AI footprint.

The echoes of 2018 are unmistakable. Google pulled out of Project Maven after employee protests over military use of its AI. The company subsequently relaxed those restrictions, and GenAI.mil represents the most direct re-engagement with Pentagon AI since that controversy.

The gap between Google's internal AI ethics debate and its external business decisions is widening. Dean's public statement and the employee letters suggest real discomfort. But the commercial incentives point in one direction: with Anthropic sidelined, the Pentagon is handing contracts to whichever frontier lab will take them without conditions. Google, OpenAI, and xAI are all taking that deal.

Sources: Google to Provide Pentagon With AI Agents for Unclassified Work - Bloomberg | Pentagon says employees can create their own 'custom AI assistants' - DefenseScoop | Google to Provide Pentagon with Gemini-powered AI agents - Engadget | Google to provide Pentagon with AI agents for unclassified work - East Bay Times | Employees at Google and OpenAI support Anthropic's Pentagon stand - TechCrunch