The Pentagon has added tailored versions of OpenAI’s ChatGPT and xAI’s Grok to a secure generative AI portal available to three million civilian and military personnel.
ChatGPT Mil and Grok for Government are now offered through GenAI.mil, a centralized service launched last year with Google Gemini. The Department of Defense says the portal provides access to commercial frontier models without routing sensitive government information through ordinary consumer products.
The additions expand a deployment that has already attracted more than 1.7 million unique users, according to the department. They also reveal different ways the Pentagon is presenting general-purpose models to staff: one focused on routine unclassified work, the other described in explicitly operational language.
A secure route to commercial models
GenAI.mil is designed to separate defense use from the data collection associated with consumer technology. The official military versions are exempt from that ordinary collection, according to the Defense Department, though TechCrunch’s report does not describe the portal’s technical controls in detail.
ChatGPT Mil emerged from OpenAI for Government and is intended to resemble the commercial ChatGPT experience. It will support chat, files, projects and custom GPTs. The department says personnel can use it for document-heavy, routine and unclassified work including administration, logistics, planning and policy. Additional features are expected later.
The stated scope emphasizes productivity rather than autonomous military action. Even so, a service used by more than half of the department’s workforce can influence how information is summarized, how plans are drafted and how institutional knowledge moves between teams.
The popularity figure comes from the Pentagon and does not show frequency of use, task quality or error rates. It does establish that secure access to commercial AI has progressed beyond a small pilot.
Grok receives a broader military pitch
The Defense Department described Grok for Government in more militaristic terms. It said the system, delivered through SpaceX’s Starshield AI, would provide immediate productivity gains, stronger knowledge continuity and more secure collaboration for warfighters.
Starshield AI is a secure satellite network built on SpaceX’s Starlink technology. The department said Grok could help personnel execute missions faster and more precisely across operational settings, giving examples that ranged from market research for acquisition teams to supply-chain management for logisticians.
Those are government projections about potential benefits, not reported performance results. The announcement does not provide tests showing that either model improves mission precision, nor does it explain how outputs will be reviewed when errors could affect procurement, logistics or planning.
The Pentagon has also signed agreements with Amazon Web Services, Microsoft, Nvidia and Reflection AI as it expands its AI capabilities. The portal therefore sits within a broader effort to combine commercial infrastructure and models with defense-specific controls.
Anthropic remains outside the portal
The new lineup highlights the absence of Anthropic’s Claude. The Trump administration designated Anthropic a supply-chain risk after the company refused to give the Pentagon unrestricted use of its systems and insisted on safety guardrails. Anthropic is challenging that designation in court.
By adding OpenAI and xAI products, the Pentagon is demonstrating that it can continue working with other frontier-model companies while the Anthropic dispute proceeds. The contrast also raises a policy question: whether vendors supplying military AI will be permitted to maintain their own limits on use.
The Defense Department frames GenAI.mil as a way to gain speed without sacrificing security. Secure hosting can reduce one category of risk—sending government data through consumer channels—but it does not by itself establish model reliability, appropriate use or accountability for generated output.
At three million potential users, those governance questions are no longer theoretical. ChatGPT Mil and Grok for Government will be judged not only by adoption, but by which tasks personnel assign to them, how human review is enforced and whether the portal’s protections match the sensitivity of the work moving through it.
