TL;DR
- New Services: ChatGPT Mil and Grok for Government are now usable on GenAI.mil by authorized defense personnel.
- Security Scope: The IL5 portal can handle controlled unclassified information, but it is separate from classified military networks.
- Model Choice: The additions give an existing Gemini-based portal different tools for document work, search, reasoning, and reusable workflows.
- Adoption Gap: GenAI.mil reported more than 1.7 million unique users, but no public figures show uptake of either new service.
Two Services Join an Existing Portal
ChatGPT Mil and Grok for Government became usable on GenAI.mil on August 31, giving authorized US service members, civilian employees, and contractors two new model services alongside Google’s Gemini. The Pentagon’s existing portal can process controlled unclassified information at Impact Level 5, but not classified material, so the additions broaden choice inside one security boundary rather than opening commercial chatbots to secret networks.
The government release changes an established service, not just a vendor roster. Gemini for Government was the first enterprise model on GenAI.mil when the portal opened in December 2025. ChatGPT Mil and Grok for Government therefore entered an already working portal rather than starting a new Pentagon AI service from zero.
The vendor relationships predate the products now on screen. In July 2025, the Pentagon awarded frontier-AI contracts to Google, OpenAI, Anthropic, and xAI. OpenAI announced plans for a GenAI.mil service in February 2026, and the department announced a Grok integration earlier in the year. Those commitments established intent, and since August 31 authorized users can now open the two new services.
Different Tools, Uneven Public Detail
A Defense Department official described Gemini as useful for search and ChatGPT for text-heavy work. Grok’s features instead center on deeper reasoning and reusable workspaces.
ChatGPT Mil opened with GPT-5.4, according to DefenseScoop’s reporting on the deployment. A person familiar with OpenAI’s work told the outlet that GPT-5.6 Terra would follow soon, so the later model was not reported as live on August 31. The current service supports ordinary chat, uploaded files, persistent projects, custom GPTs, and an Offline Search function that returns citations. The available description does not say that Offline Search browses the live public web or establish image generation for this government edition.
OpenAI says the government environment isolates user data and does not use it to train or improve the company’s public and commercial models.
Grok for Government’s disclosed features include the deep-reasoning modes labeled Auto, Fast, and Expert, along with customizable workspaces, persistent projects, and reusable playbooks. Those tools support repeated analytical work with saved context. The public launch material did not identify the underlying numeric Grok model, describe web or offline search and image or document handling, or provide a matching data-training assurance.
Impact Level 5 Covers Sensitive, Unclassified Work
Impact Level 5 is a cloud-security category for sensitive government information, including controlled unclassified information (CUI), which can include material that needs safeguards even though it is not secret. An IL5 authorization therefore lets eligible personnel use the portal for more sensitive work than a public consumer chatbot, while keeping classified data outside this environment.
That distinction matters because the Pentagon has pursued separate agreements for AI on IL6 and IL7 classified networks. Those systems use different infrastructure and approvals. Earlier promises to put leading models on unclassified and classified networks did not make every GenAI.mil service suitable for secret or top-secret material.
Access also remains governed by department and service rules. GenAI.mil’s IL5 environment is for unclassified work and Navy guidance permits CUI while excluding personally identifiable and protected health information. That service-specific guidance illustrates why an IL5 label is a security ceiling, not permission to paste every kind of sensitive record into any model.
Availability Is Not Adoption
The roughly three million people offered access on GenAI.mil and the more than 1.7 million unique users reported on August 31 measure different things. The first figure describes eligibility; the second counts unique users across GenAI.mil. Neither measures ChatGPT Mil or Grok for Government use after their release specifically
Model-specific usage, task categories, and measured outcomes would show whether personnel are using the additional choices and for what work. Until those figures are available, the supported change is access to ChatGPT Mil and Grok for Government, not an already measured improvement in defense work.
