Summary
- The US Department of Defense has added OpenAI's ChatGPT Mil and xAI's Grok for Government to its GenAI.mil portal.
- The department says more than 1.7 million of its 3 million personnel have already signed up for GenAI.mil.
- Anthropic's Claude remains excluded after being designated a "supply chain risk" following a dispute over safety guardrails.
ChatGPT and Grok join the Pentagon's portal side by side
The US Department of Defense says it has added OpenAI's ChatGPT Mil and Grok for Government — built by xAI and now operated under SpaceX — to GenAI.mil. Launched last year with Google Gemini as its headline offering, GenAI.mil is the department's central, secure portal designed to give defense personnel access to frontier AI models without the kind of data collection that's hard to avoid when using commercial tech products. According to the department, more than 1.7 million of its 3 million civilian and military personnel have already signed up for GenAI.mil.
ChatGPT Mil targets paperwork, Grok targets mission execution
ChatGPT Mil, part of OpenAI's government program, carries over the chat, file, project, and custom GPT features familiar to everyday ChatGPT users. It's aimed at unclassified, document-heavy, repetitive work — administration, logistics, planning, and policy — and the department says more capabilities are coming.
Grok, by contrast, was introduced in much more military terms. Delivered through SpaceX's secure satellite network, Starshield, Grok for Government was described in the Pentagon's press release as designed to "provide the warfighter with immediate productivity gains." The department added that the tool is meant to help personnel act faster and more precisely — from market research for acquisitions officers to supply chain management for logistics teams.
Why Claude is missing
With two new tools joining the portal, what stands out is the absence of Anthropic's Claude. When the Pentagon asked for unrestricted use of Claude, Anthropic declined, saying it would keep certain safety guardrails in place. The Trump administration then designated Anthropic a "supply chain risk" company. Anthropic is currently fighting that designation in court, arguing it's unwarranted.
To put it plainly: when the Defense Department asked to use Claude in military operations without restriction, Anthropic refused to strip out its safety guardrails. Citing that refusal, the Trump administration designated Anthropic a "supply chain risk," a label Anthropic is now challenging in court as unjustified. Meanwhile, the department turned to OpenAI and xAI instead.
The Pentagon's AI procurement keeps expanding
The Defense Department has already signed deals with a range of tech companies — including AWS, Microsoft, NVIDIA, and Reflection AI — to build out its AI capabilities. OpenAI, for its part, has been expanding into defense and security work since releasing GPT-5.6-Cyber, a model built for cybersecurity tasks, back in August — a move we covered in detail in OpenAI unveils GPT-5.6-Cyber, a dedicated cybersecurity model. This GenAI.mil expansion fits the same pattern: the department appears to be spreading its bets across multiple frontier AI companies at once, reducing reliance on any single vendor.
Editor's take
The real story here isn't that ChatGPT and Grok got added — it's why Claude didn't. The Pentagon labeled the company that insisted on safety guardrails a "supply chain risk," then filled its portal with companies that didn't push back on guardrails. That's not just a procurement decision. It's a signal about which side of the divide — vendors who hold the line on safety, versus vendors who cooperate without conditions — the department is choosing to favor in the military AI market.
For OpenAI and xAI, this is a genuine business win. Landing a spot in a portal serving 3 million government users means locking in steady demand that's hard to come by in the commercial market. xAI in particular has been building its presence in government and security work since being absorbed into SpaceX, and framing Grok in military-operations language in this press release fits squarely into that push.
For companies watching from outside the US, this episode points to two lessons. First, if you want to supply AI into public-sector or defense markets, you need to decide in advance where you'll draw the line between safety guardrails and customer demands. Hold firm like Anthropic did, and you might lose the contract. Chase the contract without any guardrails, and you might face accountability problems down the road. Second, government customers clearly favor a multi-vendor structure — spreading contracts across several suppliers rather than betting everything on one. Running several frontier models side by side, rather than committing to a single vendor, is becoming the default approach in government AI procurement.
Two things are worth watching over the coming weeks: how Anthropic's lawsuit over its "supply chain risk" designation plays out, and how much of GenAI.mil's 1.7 million registered users actually end up using ChatGPT Mil and Grok. Both numbers will shape where the Pentagon's next round of AI procurement deals goes.
Comments