The Pentagon Blacklisted Anthropic for Safety Limits. A Judge Called That Illegal. ChatGPT and Grok Are Live Anyway.

On August 31, the Department of Defense opened ChatGPT Mil and Grok for Government to 3 million military personnel, civilian employees, and contractors through its GenAI.mil portal. Claude was not on the list.

The timing was pointed. Three days earlier, Federal Judge Rita Lin ruled that the Pentagon's designation of Anthropic as a "supply-chain risk" — the legal mechanism that had kept Claude off the platform — was unconstitutional. Judge Lin found that the designation was unlawful retaliation after Anthropic publicly criticized the government's use of AI. She called the evidence of a national-security risk "slim" and concluded that the government had punished a company for speaking up.

The ruling invalidated the designation, but it did not put Claude on GenAI.mil. When the Pentagon expanded the portal, Google Gemini, ChatGPT Mil, and Grok for Government were its three model offerings; Anthropic remained outside the contract.

A Platform Built for Scale

GenAI.mil launched last year with Google's Gemini as its first model offering. The idea was straightforward: rather than routing sensitive government work through commercial consumer apps, the Pentagon would offer a centralized portal with government data controls. All three models now on the platform have Impact Level 5 authorization, the tier for sensitive unclassified and controlled unclassified information that does not reach the classified threshold.

By June 2026, GenAI.mil had 1.5 million users. By August 31, it had reached 1.7 million, or 57 percent of the eligible workforce. The new deployment widens potential access to 3 million people across the department.

ChatGPT Mil, built on GPT-5.4 Terra, is positioned for document-heavy unclassified work such as logistics planning, policy analysis, supply-chain management, and routine administration. Grok for Government adds xAI's adaptive reasoning modes, customizable workspaces, and reusable playbooks. A DoD official described the rationale simply: a diverse suite of AI capabilities should help warfighters act with confidence.

That vendor diversity is also procurement leverage. Three frontier-model providers on a shared secure platform give the department more ability to compare systems, avoid lock-in, and negotiate at scale.

The Guardrail Dispute

The Anthropic exclusion grew out of a dispute over the terms for putting Claude on GenAI.mil. Anthropic held to usage guardrails that would prevent Claude from being used to plan mass surveillance or assist with lethal autonomous-weapons decisions.

Defense Secretary Pete Hegseth argued that the U.S. government could not allow a private company to dictate how the military uses its tools. In February 2026, the Pentagon designated Anthropic a supply-chain risk under an authority usually associated with companies tied to foreign adversaries. The designation effectively barred Pentagon components and a broad network of defense contractors from working with Anthropic products.

Anthropic sued. Judge Lin's August 28 ruling held that the designation was not a legitimate national-security determination, but retaliation for protected speech. She also found that Anthropic had been denied a meaningful opportunity to contest the designation before it took effect, a Fifth Amendment problem alongside the First Amendment violation. The designation was struck down.

The commercial dispute, however, has not disappeared. A court can invalidate a designation; it cannot require the Pentagon and Anthropic to agree on a contract. That is why a legal win for Anthropic and its absence from the portal can coexist.

What the Market Just Learned

The GenAI.mil expansion puts a sharp edge on a question AI companies have been circling for two years: what happens when safety commitments made to researchers, regulators, and users conflict with the requirements of a customer with uniquely powerful demands?

OpenAI and xAI reached arrangements with the Defense Department that Anthropic did not. That is not a simple moral ranking; it reflects different choices about where to draw usage boundaries and who has authority to draw them. For the DoD, a frontier model that can operate within military doctrine is more immediately useful than one that retains restrictions the department rejects.

Anthropic's court victory is real, but it does not turn into access by itself. Its guardrails may become a competitive qualification if international frameworks, including rules for high-risk AI systems and emerging NATO guidance, move toward stronger restrictions on lethal autonomous systems. If U.S. defense AI policy keeps moving in the other direction, the market signal will be different.

For now, ChatGPT Mil and Grok are joining Gemini on government workstations. Claude is no longer formally blacklisted under the struck-down designation, but it is still waiting for a contract that neither the court nor the Pentagon is required to provide.

Sources

DefenseScoop — Grok, ChatGPT added to GenAI.mil: https://defensescoop.com/2026/08/31/grok-chatgpt-added-to-genai-mil/

TechCrunch — The Pentagon now has its own version of ChatGPT and Grok: https://techcrunch.com/2026/08/31/the-pentagon-now-has-its-own-version-of-chatgpt-and-grok/

TechCrunch — Anthropic gets its first court win over the Pentagon's supply-chain-risk label: https://techcrunch.com/2026/08/28/anthropic-gets-its-first-court-win-over-the-pentagons-supply-chain-risk-label/

CNN — Judge rules Anthropic supply-chain-risk designation unlawful: https://www.cnn.com/2026/08/27/tech/anthropic-pentagon-supply-chain-risk-unlawful-hnk

Nextgov/FCW — Judge rules Anthropic supply-chain-risk designation was illegal and baseless: https://www.nextgov.com/artificial-intelligence/2026/08/judge-rules-anthropic-supply-chain-risk-designation-was-illegal-and-baseless/415698/

NOTUS — Pentagon broadens AI tools with ChatGPT and Grok: https://www.notus.org/defense/pentagon-broadens-ai-tools-chatgpt-grok