Pentagon Tests OpenAI and Google AI Models to Potentially Replace Claude in Defense

# Pentagon Tests OpenAI and Google AI Models to Potentially Replace Claude in Defense

The U.S. Department of Defense is running a live bake-off between frontier AI models from OpenAI, Google, and xAI after blacklisting Anthropic — and the outcome could reshape how the world's largest military bureaucracy integrates artificial intelligence for years to come.

Twenty-five designated military "power users" began testing rival AI systems on March 1, just three days after Defense Secretary Pete Hegseth formally designated Anthropic a supply-chain risk, effectively banning its Claude models from Pentagon networks. The evaluation is running through GenAI.mil, the enterprise generative AI platform that now serves over 1.3 million Defense Department personnel and has processed tens of millions of prompts since its December 2025 launch.

The stakes are enormous. Anthropic had secured a contract worth up to $200 million in July 2025 to integrate Claude into classified military networks at Impact Level 6 and Impact Level 7 — the Pentagon's most sensitive cloud environments, reserved for classified and top-secret national security data. Claude was, at the time, the only frontier AI model operating inside those classified workflows, deployed largely through a partnership with Palantir's Maven platform.

That arrangement unraveled when Pentagon officials pushed for unrestricted use of the technology in areas including national security surveillance and operational planning. Anthropic refused to lift safety restrictions that prevent its AI from being used for mass surveillance or lethal autonomous weapons systems. The standoff escalated in January 2026 when reports emerged that Claude had been used for intelligence analysis related to Iran, and by late February, Hegseth had labeled Anthropic a supply-chain risk — a designation normally reserved for foreign adversaries.

Anthropic has challenged the designation in federal court, arguing it could cost the company billions in revenue. The litigation remains active.

Eight Companies, Zero Anthropic

On May 1, the Pentagon announced formal agreements with eight technology companies to deploy frontier AI on its classified networks: SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon Web Services, and Oracle. Anthropic was conspicuously absent.

"It's irresponsible to be reliant on any one partner," Emil Michael, the Pentagon's CTO and Undersecretary of Defense for Research and Engineering, told CNBC. "We learned that that one partner didn't really want to work with us in the way we wanted to work with them. We went out and made sure that we had multiple different providers — both open source and the proprietary model companies — and got them to agree to sign up to work with us on classified networks."

The multi-vendor approach reflects a deliberate pivot away from single-source dependency. Pentagon AI chief Cameron Stanley confirmed to CNBC that the department is expanding its use of Google's Gemini model while simultaneously working with OpenAI and others to modernize wartime capabilities. The goal is to prevent "vendor lock" — a sign of how central AI has become to military infrastructure.

Internal Resistance at Google

The Pentagon's courtship of new AI partners has not been frictionless. More than 700 Google employees signed a letter to CEO Sundar Pichai in late April urging the company to reject classified military work, warning that such arrangements could allow Google's technology to be used in ways they described as inhumane or extremely harmful. The protest echoed Google's 2018 Project Maven controversy, when employee backlash forced the company to abandon an earlier Pentagon AI contract.

This time, the deal went through anyway.

What the Experts Say

Lauren Kahn, a senior research analyst at Georgetown's Center for Security and Emerging Technology and a former DOD policy advisor, called the multi-vendor expansion "a step in the right direction" and "long overdue."

"These models all have different strengths and are evolving and updating at lightspeed, so it's important to have access to a variety and let Pentagon users shop around," Kahn told DefenseScoop. She noted that Anthropic had been the first company operating on the classified side, effectively paving the way for these other companies to follow.

Kahn emphasized that the real challenge lies ahead: training the 1.3 million-plus GenAI.mil users to guard against automation bias and understand the limitations of the systems they are using. Having multiple models, she noted, actually accelerates that learning because users can directly compare responses, accuracy, and speed.

What to Watch

Three dynamics will shape this story in the months ahead.

First, Anthropic's lawsuit. If the courts overturn the supply-chain risk designation, the Pentagon could face pressure to reinstate Claude alongside its new vendor roster — creating an even more competitive multi-model environment.

Second, the evaluation results. The 25-person test panel is comparing models on workflows that previously relied on Claude. Initial reports suggest the models respond differently to the same prompts, which means the Pentagon may end up deploying different AI systems for different mission sets rather than picking a single winner.

Third, the ethics fault line. Anthropic's refusal to remove safety guardrails — and the Pentagon's willingness to find vendors who would — sets a precedent for how the defense establishment negotiates AI safety boundaries. OpenAI has already renegotiated its terms to permit unrestricted "lawful use" of its technologies by the military, a sharp departure from its original nonprofit charter. Contractors who had integrated Anthropic's technology have been given a six-month window to find alternatives.

The Pentagon's message is clear: in the race to become an AI-first fighting force, it wants options — and it will not let any single company's ethical red lines dictate the pace.