The Defense Department put OpenAI’s ChatGPT Mil and Starshield AI’s Grok for Government onto GenAI.mil on Monday, giving more than 3 million military personnel, civilian staff and contractors two more commercial models for sensitive unclassified work. The portal launched in December 2025 with Google’s Gemini for Government, and more than 1.7 million people have used it since.
Both new tools cleared Impact Level 5, the highest authorization for systems handling non-public, sensitive unclassified data. ChatGPT Mil is aimed at document-heavy work: planning, policy, logistics and administration. Its core is “chat, files, projects, and custom GPTs, with additional features sequenced over time,” according to the department. Personnel currently reach GPT-5.4 Terra, with GPT-5.6 Terra to follow. Grok for Government carries adaptive reasoning modes, persistent projects and reusable “playbooks,” and the department says it can be used for acquisition research and supply chain management.
Anthropic was supposed to be on the platform too. When the platform was unveiled, officials named Claude alongside ChatGPT and Grok as models that would follow Gemini. Instead Anthropic held out for contract terms barring the use of Claude for mass surveillance of Americans and for lethal autonomous weapons, the department demanded unrestricted use for what it deems lawful purposes, and in February it designated the company a supply chain risk and ordered its products purged from national security systems. A federal judge vacated that designation last week as unlawful retaliation. Government lawyers have said the purge will finish by the end of September.
An official who asked to be identified as a War Department official said the department “will continue to build an architecture that prevents AI vendor lock and ensures long-term flexibility for the Joint Force,” and that it is “moving at wartime speed.” President Trump signed an executive order last year rebranding the Defense Department as the Department of War; changing the name in law requires Congress.
OpenAI reached the terms Anthropic would not. Its agreement permits lawful military use while keeping limits on domestic surveillance, autonomous weapons and certain high-stakes decisions made without a human, enforced contractually and technically by retaining control of its safety systems and delivering models through the cloud rather than onto weapons or devices. Google took a comparable route in April.
What Monday’s announcement does not say is whether those OpenAI limits carry over to ChatGPT Mil. The department describes the tool as suitable for planning without specifying whether that means operational or military planning, and says nothing about intelligence analysis, targeting or decisions involving force. Asked directly, a Pentagon official said only that everything on GenAI.mil is certified for controlled unclassified information at Impact Level 5.
The expansion is running alongside unresolved security questions. The department’s counterintelligence agency warned in June that unauthorized “shadow AI” tools inside the workforce could leak data, and Congress has ordered an assessment of cybersecurity risks from both approved and unapproved AI software. A separate effort announced in May would put commercial models on classified networks at Impact Levels 6 and 7, for situational awareness and warfighter decision-making.
Joe Larson, a former senior Pentagon official who now runs OpenAI’s government business, said the integration pairs “access to powerful tools with clear safeguards, meaningful human judgment, and accountability for high-stakes decisions.” xAI and SpaceX did not respond to requests for comment.
Sources: DefenseScoop, NOTUS
–
By the Control Plane Editorial Team