Anthropic’s Claude is the one frontier model missing from the Pentagon’s AI portal

The Pentagon on Monday pushed military versions of OpenAI’s ChatGPT and xAI’s Grok to its GenAI.mil portal.

The two tools reach 3 million defense personnel. Anthropic’s Claude is locked out in a court fight over the safety limits it wanted.

1.7 million users signed on in nine months

GenAI.mil is the Department of War’s one-stop shop for commercial AI.

Staff can deploy frontier models without passing sensitive data through consumer apps, and military versions omit the data collection built into most consumer products.

Both tools received accreditation for Controlled Unclassified Information at Impact Level 5, the tier that covers sensitive government work that is not classified. That clearance allows them to deal with it at scale.

It launched about nine months ago with Google’s Gemini as its first model. The department says it has enrolled more than 1.7 million users out of a workforce of more than 3 million.

With the addition of Grok and ChatGPT, the portal becomes a multi-model marketplace. The Pentagon framed that as a way to avoid getting locked into one supplier.

ChatGPT Mil is the result of a government program at OpenAI and an enterprise partnership forged by the two sides in 2025. It starts with chat, files, projects, and custom GPTs for document-heavy unclassified work across planning, policy, logistics, and administration.

The department points to Starshield AI. The secure-satellite unit of SpaceX runs on Starlink technology.

Its release lists deep-thinking inference, three modes of reasoning called Auto, Fast, and Expert, and reusable “playbooks” meant to capture institutional knowledge. Applications range from acquisition market research to supply chain work for logisticians.

The department said Grok enables forces to “execute missions faster and with greater precision.” The two releases are different, with the Grok one describing the execution of the mission and the ChatGPT one describing the desk work.

Judge Rita Lin saw a First Amendment problem

Although Anthropic won its federal court ruling to lift the national security blacklist, its Claude models remain excluded from the GenAI.mil portal. A parallel challenge is proceeding in the D.C. Circuit Court.

Late in February, Defense Secretary Pete Hegseth called Anthropic a Supply-Chain Risk to National Security, Cryptopolitan reported. This followed President Donald Trump ordering federal agencies to stop using the company’s technology.

Anthropic, which inked a $200 million Pentagon deal in July, wanted written assurances that its models would not be used in fully autonomous weaponry or mass domestic surveillance of Americans.

The Pentagon pushed for unrestricted access for all legal purposes and set a deadline of February 27 that was not met.

Anthropic called the label “unprecedented,” saying it’s normally reserved for foreign adversaries and vowed to fight it.

The company pointed out its history as the first frontier lab to operate models on classified U.S. networks and to support warfighters since June 2024. It pointed to a federal procurement law, saying Hegseth lacks the authority to prevent contractors from using Claude for non-defense work.

Although U.S. District Judge Rita Lin expressed First Amendment concerns during an early March hearing, she issued a final 59-page ruling on August 27, 2026, which officially struck down the Pentagon’s supply-chain risk designation.

Lin has not issued a final ruling. She said the government responded harshly after Anthropic went public with the dispute and said the crackdown “doesn’t seem to be really tailored to a stated national security concern.”

Anthropic’s lawyer, Michael Mongan, told the court that the designation had never been applied to a U.S. company before. The company is asking the court to lift the supply chain risk.

Defense staff currently gain access to OpenAI and xAI tools, whereas Anthropic remains locked out of the portal pending the outcome of the D.C. circuit court appeal.