Pentagon adds ChatGPT and Grok to its military AI hub as court rules Anthropic blacklist unlawful

The Pentagon's in-house artificial intelligence hub grew more crowded this week, and the empty seat at the table has become just as newsworthy as the new arrivals.

On August 31, the Department of War said it had added OpenAI's "ChatGPT Mil" and Starshield AI's "Grok for Government" to GenAI.mil, the department's secure generative-AI portal, putting both tools alongside Google's Gemini for a workforce of more than three million military and civilian personnel. The rollout, confirmed in separate statements reported by DefenseScoop and TechCrunch, makes the department one of the few federal agencies to offer staff a menu of competing frontier chatbots rather than a single vendor's product.

Conspicuously missing from that menu is Claude, the model built by Anthropic. Its absence is not incidental: it is the residue of a year-long fight between the AI company and the department over whether Claude could be used for autonomous weapons and domestic surveillance, a fight that reached a new turn just three days before the ChatGPT and Grok launch, when a federal judge ruled that the Pentagon's attempt to blacklist Anthropic had been unlawful.

The numbers behind the rollout

GenAI.mil has onboarded more than 1.7 million unique users out of the department's roughly three million personnel since Gemini became the platform's first offering in December 2025, according to figures cited in the department's own announcements and repeated by DefenseScoop and TechCrunch. Both ChatGPT Mil and Grok for Government have cleared Impact Level 5, the Pentagon's accreditation standard for handling Controlled Unclassified Information, putting them on the same security footing as Gemini.

ChatGPT Mil is built around chat, file uploads, shared "projects" and custom GPTs, and is pitched at document-heavy, unclassified work such as logistics planning, policy drafting and administrative tasks; the department has said personnel currently have access to OpenAI's GPT-5.4 Terra model, with the newer GPT-5.6 Terra to follow. Grok for Government, run through Elon Musk's Starshield AI, adds "deep-thinking" inference, adjustable reasoning modes and reusable "playbooks" meant to capture institutional knowledge across a unit. xAI's broader push into federal work was cemented last year through a General Services Administration agreement offering Grok models to federal agencies for 42 cents through March 2027, one of the cheapest terms GSA has struck under its OneGov AI purchasing program.

The multi-vendor approach traces back to a falling-out with Anthropic that Pentagon officials have cited publicly as a cautionary tale. Emil Michael, the department's chief technology officer, told reporters earlier this year that the experience of one AI company that "didn't really want to work with us in the way we wanted to work with them" reinforced the case for spreading the department's AI relationships across "multiple different providers," including cloud and infrastructure vendors such as Microsoft and Amazon Web Services alongside model makers. A War Department official quoted in coverage of this week's launch put the same logic more bluntly, saying access to "a diverse suite of AI capabilities from across the resilient American technology stack" would give troops the tools they need without depending on any single supplier.

That vendor Michael alluded to was Anthropic. Negotiations over bringing Claude onto GenAI.mil broke down last year after Anthropic refused to grant the department "any lawful use" of the model, insisting instead on carve-outs barring Claude from fully autonomous lethal weapons and from mass domestic surveillance of Americans. In a statement posted in February, Anthropic said the impasse followed "months of negotiations that reached an impasse over two exceptions we requested to the lawful use of our AI model, Claude." Defense Secretary Pete Hegseth responded by directing the department to designate Anthropic a "supply chain risk" to national security, a formal letter the company received on March 4 and described in a follow-up post as applying narrowly to Claude's use directly within department contracts rather than to its commercial customers generally.

A court rules the blacklist unlawful

Anthropic sued in March, arguing the designation amounted to unconstitutional retaliation for the company's public criticism of Pentagon AI policy and denied it due process. The case, Anthropic PBC v. U.S. Department of War, landed before U.S. District Judge Rita Lin in the Northern District of California, who granted Anthropic a preliminary injunction in March. After a federal appeals court declined to keep that injunction in place in April, the underlying merits fight continued, culminating in a summary-judgment ruling issued the night of August 27.

Judge Lin found that the department's actions violated Anthropic's First Amendment rights and its due-process protections under the Fifth Amendment, concluding that officials had sought to "impose sweeping penalties based principally on Anthropic's critique" of the department's AI ambitions rather than on any genuine security threat.

"The empty invocation of national security is not a blank check to punish and retaliate against government critics."

An Anthropic spokesperson said the company "welcome[s] the court's ruling that this supply chain risk designation was unlawful," adding that it "remain[s] focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology." The Pentagon did not immediately comment on the ruling, according to NPR's reporting, and is widely expected to appeal.

Who is affected

For the roughly three million people eligible for GenAI.mil accounts — uniformed service members, civilian employees and, increasingly, contractors working alongside them — the practical effect of this week's launch is more choice: a document-drafting task that once defaulted to Gemini can now be routed to ChatGPT Mil or Grok for Government instead, each accredited to handle the same sensitive but unclassified material. For defense contractors more broadly, the supply-chain-risk designation had barred use of Claude in Pentagon-related work regardless of the court fight's outcome for ordinary commercial customers, a restriction Judge Lin's ruling now unwinds, at least for as long as it survives on appeal.

For OpenAI and xAI, the launch cements paid footholds inside one of the world's largest employers, following separate, smaller-scale entrants such as Nvidia, Reflection AI and the cloud providers that TechCrunch has reported also hold Pentagon AI deployment agreements. For Anthropic, the ruling is a legal win that does not, on its own, restore Claude to GenAI.mil; the company has said its position on autonomous weapons and surveillance is unchanged, and a second Anthropic suit against the department remains pending in Washington, D.C.

What happens next

The department has not said whether Thursday's ruling will prompt a renewed attempt to bring Claude onto GenAI.mil, and Anthropic's own statements suggest it has no plans to soften the restrictions that triggered the original dispute. An appeal of Judge Lin's summary-judgment order is expected, which would extend a legal fight now well into its second year and leave the supply-chain-risk designation's ultimate fate unresolved even as ChatGPT Mil and Grok for Government go live for millions of users this week. In the meantime, GenAI.mil's next model upgrade — OpenAI's GPT-5.6 Terra — is expected to reach department users in the coming weeks, department officials have said, continuing a rollout that began with a single vendor nine months ago and now spans three.