The U.S. Department of Defense has activated OpenAI’s ChatGPT Mil and the Grok for Government software operated by xAI subsidiary Starshield on its internal platform GenAI.mil since September 1, 2026, accessible to around three million civilian and military employees. Anthropic’s Claude is still missing – although a federal judge dismissed the company’s classification as a security risk as unlawful at the end of August.
GenAI.mil gets two new tools for millions of employees
GenAI.mil launched in December 2025 with Google’s Gemini as a central, secure portal for commercial AI models. Sensitive government data is not supposed to flow through ordinary consumer channels there, and the services are exempt from the usual data collection of the civilian versions, as TechCrunch reports. So far, 1.7 of the department’s 3 million employees have registered on the platform. ChatGPT Mil is aimed primarily at document-heavy, unclassified everyday work such as administration, logistics, planning, and policy, offering chats, file storage, projects, and custom GPTs. Grok for Government, according to Pentagon statements, brings multi-step reasoning with Auto, Fast, and Expert modes, persistent projects, and reusable “playbooks” for market analysis and supply-chain management. Both new tools carry the Impact Level 5 security clearance, which permits handling controlled but unclassified information. That classification, the Pentagon notes, says nothing about a model’s suitability for a specific task. Related tools like OpenAI’s agent ChatGPT Work target very similar tasks in the civilian market.
A dispute over safeguards led to the spring ban
The exclusion stems from a conflict that began long before the current rollout. In July 2025, the Pentagon awarded contracts to four providers – Anthropic, Google, OpenAI, and xAI – under identical terms for up to $200 million each for “agentic” AI capabilities. The department explicitly wanted to ensure competition and avoid dependence on a single provider. When it came to the specific integration of Claude into GenAI.mil, the department demanded unrestricted access to the model for any lawful purpose. Anthropic insisted instead on contractual commitments that its technology would not be used for fully autonomous weapon systems or domestic mass surveillance. In February 2026, President Donald Trump and Defense Secretary Pete Hegseth publicly labeled the company a security risk. In March, the formal classification as a “supply chain risk” under U.S. law followed – effectively excluding Anthropic from government contracts. Anthropic CEO Dario Amodei held his ground and went to court, initially failing in April in an attempt to block the classification through an emergency motion.
A court ruling overturns the classification, the exclusion remains
On August 27, 2026, Federal Judge Rita Lin ruled for Anthropic on the central claim. The supply-chain-risk classification violated the right to free speech and the right to due process, as Fortune quotes from the ruling. The department had punished Anthropic out of arrogance, the ruling stated, and the mere invocation of national security was no blank check to punish government critics. The judge formally vacated Hegseth’s classification and barred its further enforcement. The rollout of ChatGPT Mil and Grok a few days later was unaffected by the ruling. Claude appears in none of the announcements around GenAI.mil so far, and the department has not said whether or when it might admit Claude retroactively. Anthropic’s separate lawsuit in a Washington court continues independently, the company says. It had already been apparent from the ruling itself that the contract exclusion would initially persist despite the legal win. That assessment has now been confirmed by the launch of ChatGPT Mil and Grok.
What will matter is whether the Pentagon revises its stance after the courtroom defeat, or whether Anthropic stays permanently excluded from one of the U.S. government’s largest AI procurement programs despite the ruling. The case also shows how directly political fights over weapons use and surveillance by AI models can decide market share in a billion-dollar contracting business. A similar pattern already showed up in the distribution of AI spending in the U.S. Congress, where ChatGPT likewise captured a far larger share than Claude.
