Google in Talks With Pentagon to Deploy Gemini Across All Classification Levels

Google Offers Pentagon Full Access to Gemini

Google is in active discussions with the U.S. Department of Defense about deploying its Gemini AI model across Pentagon systems at every classification level — from unclassified to top secret. The Information reported the development, citing people familiar with the matter.

The potential agreement would allow the DoD to use Gemini "for all lawful purposes." During negotiations, Google proposed additional provisions that would prohibit the use of Gemini for mass surveillance of citizens or the development of fully autonomous weapons systems.

A Pentagon spokesperson declined to comment directly on the talks with Google but emphasized that the department intends to continue adopting cutting-edge AI technologies through close collaboration with the private sector across all classification levels.

Why This Matters

A deal between Google and the Pentagon would represent a fundamental shift in the tech giant's relationship with the military. Signing a contract that includes classified environments would mean full-scale integration of commercial AI into core defense operations — far beyond the limited cooperation of the past.

The negotiations unfold against the backdrop of a major clash between the Pentagon and the AI industry over ethical guardrails — a conflict that already resulted in litigation with Anthropic and a presidential ban on the startup's technology across all federal agencies.

The Anthropic Precedent: When Ethics Triggered a Breakup

Google's talks come in the wake of a high-profile confrontation between the Pentagon and AI startup Anthropic. In July 2025, the DoD awarded contracts worth up to $200 million to Anthropic, Google, OpenAI, and xAI for developing AI solutions in the security domain. Among all contractors, only Anthropic's tools were integrated into classified environments due to their quality.

By January 2026, however, the WSJ reported that the agreement was at risk of collapse. Anthropic's strict ethical policies — which forbid using Claude for mass surveillance and autonomous lethal operations — frustrated officials. Defense Secretary Pete Hegseth, commenting on the partnership with xAI, stated that the department "will not use models that don't allow you to fight wars."

The crisis escalated in February 2026 when the U.S. military used Claude in an operation to capture Venezuelan President Nicolás Maduro. Anthropic CEO Dario Amodei declared that his company would rather walk away from the Pentagon than allow its technology to be used in ways that could "undermine rather than protect democratic values." He confirmed that the core issues were the potential use of tools like Claude for "mass domestic surveillance" and "fully autonomous weapons."

The conflict escalated into litigation and a ban by President Donald Trump on Anthropic's technology across all federal agencies. Paradoxically, Anthropic's principled stance boosted its popularity among users. OpenAI, meanwhile, faced criticism for its own defense agreement — Sam Altman later called the decision hasty.

From Project Maven to Classified Networks: Google's Evolution

The current negotiations echo events from 2018, when Google pulled out of Project Maven — a program that used AI to analyze drone surveillance footage. At the time, more than 3,000 employees signed an open letter demanding that leadership halt AI development for combat drones.

Since then, Google has methodically rebuilt its government relationships. In 2022, the company established a dedicated division for public sector work. In 2024, it signed an AI cooperation agreement with the DoD, though limited to unclassified applications. In early 2025, Google removed a clause from its internal AI principles that explicitly prohibited the use of its technology in "weapons and surveillance systems."

Internal Resistance Persists

Google's expanding defense work continues to face pushback from within. In April 2024, reports emerged that Google was providing cloud computing and AI services to the Israeli Ministry of Defense. Roughly 200 DeepMind employees — Google's premier AI research division — filed a formal protest, calling for the company to terminate agreements with defense agencies.

"Any involvement in military and weapons production negatively impacts our position as a leader in ethical and responsible AI and contradicts our mission and stated AI principles," the document stated, signed by approximately 5% of the division's staff.

The Pentagon Demands AI Without Guardrails

In February, reports surfaced that the Pentagon has been urging leading AI companies to make their tools available in classified networks without the standard safety restrictions typically applied to users. These networks handle sensitive tasks ranging from operational planning to weapons targeting.

The military wants to leverage AI's ability to synthesize information and support decision-making. Yet neural networks are prone to errors and hallucinations — mistakes that could prove fatal in combat scenarios. The DoD aims to make AI a central element of its strategy to connect sensors and strike systems in combat operations, accelerating the decision-making cycle.

Google's proposal to include contractual restrictions on surveillance and autonomous weapons appears to be an attempt to strike a balance between Pentagon demands and employee concerns — the very equilibrium that Anthropic could not maintain.