The Department of War (the renamed Department of Defense) rolled out custom versions of ChatGPT and Grok on its internal AI platform this week. More than three million civilian and military personnel now have access to tools built for what the department calls “warfighter needs.”
Google's Gemini was already part of that ecosystem. Anthropic's Claude isn't — and given the Pentagon's very public fight with Anthropic over how AI can be used, that absence is difficult to treat as incidental.
What just launched
GenAI.mil is the Pentagon's secure enterprise AI platform, built to give personnel access to frontier models inside authorized government cloud infrastructure rather than sending government work through ordinary consumer services. Google launched Gemini for Government on the platform in December 2025, making it the first enterprise AI tool available through GenAI.mil.
On August 31, two more tools joined it: ChatGPT Mil, built through OpenAI's government program, and Grok for Government, delivered under SpaceX's Starshield AI brand. The Pentagon says more than 1.7 million unique users have signed up since GenAI.mil launched nine months ago, out of more than three million personnel — roughly 57% of the department.
That's a striking adoption rate for a platform operating across one of the world's largest organizations.
ChatGPT Mil: the familiar option
ChatGPT Mil looks deliberately close to the product most developers already know: chat, file uploads, projects, and custom GPTs. The Pentagon says it is intended for document-heavy unclassified work including planning, policy, logistics, administration, procurement, research, and mission support.
OpenAI says the deployment runs in authorized government cloud infrastructure, with model- and platform-level safeguards, and that data processed on GenAI.mil remains isolated from OpenAI's public and commercial model-training systems.
More capabilities are scheduled to arrive over time. For now, the pitch is relatively straightforward: take the familiar ChatGPT experience and put it inside a government-controlled environment.
Grok for Government: a different pitch
Grok's positioning is more operational. The Pentagon says Starshield AI's Grok for Government is accredited for Controlled Unclassified Information at Impact Level 5, and includes deep-thinking inference, three reasoning modes — Auto, Fast, and Expert — along with customizable workspaces, persistent projects, and reusable “playbooks” designed to preserve institutional knowledge as personnel rotate.
Its stated use cases range from market research for acquisition professionals to supply-chain management for logisticians. In other words, the department is pitching Grok not simply as a chatbot, but as an enterprise reasoning and workflow tool that can be embedded into recurring military and administrative processes.
The tool that's missing
Here's what makes this more interesting than another government AI launch: Claude isn't on GenAI.mil.
Gemini, ChatGPT, and Grok are now part of the platform's frontier-model lineup. Anthropic isn't.
The Pentagon has not publicly said that its dispute with Anthropic is the specific reason Claude is absent from GenAI.mil. But the timing matters: the department spent much of 2026 fighting Anthropic over precisely the restrictions that the company wanted attached to government use of Claude.
Why Anthropic said no
Back in February 2026, Anthropic and Defense Secretary Pete Hegseth were in a public standoff. Anthropic said negotiations had reached an impasse over two narrow restrictions: using Claude for mass domestic surveillance of Americans and using it in fully autonomous weapons. Anthropic argued that today's frontier models were not reliable enough for fully autonomous lethal weapons, while also saying it supported lawful national-security uses outside those two exceptions.
Hegseth's position was fundamentally different. The Pentagon argued that the department, not a private AI company, should determine what lawful military applications its technology could support. On February 27, Hegseth directed the department to designate Anthropic a supply-chain risk to national security, following President Trump's order that federal agencies stop using Anthropic's technology.
Anthropic sued in California and Washington, D.C.
The litigation then split into two very different tracks. In California, Judge Rita Lin issued a preliminary injunction on March 26 blocking the government from implementing the presidential directive and the supply-chain designation while the case proceeded. The order explicitly said it did not prevent the Department of War from moving to other AI providers, provided those actions complied with applicable law.
The parallel D.C. litigation initially went the other way: the D.C. Circuit declined to block the Pentagon's action while that case continued.
Then, on August 27, the California case produced the much bigger development. Judge Lin ruled on the merits of Anthropic's challenge, granting Anthropic summary judgment on key claims against the Department of War and finding the challenged designation unlawful. The court also rejected the government's request to pause the effect of the ruling.
That makes the August decision Anthropic's biggest court victory yet in the dispute — but not its first court victory.
The legal backdrop is also more nuanced than “the Pentagon allows AI to kill without humans.” A 2023 DoD directive permits autonomous and semi-autonomous weapon systems, including systems with automated functions for acquiring, tracking, identifying, prioritizing, and engaging targets. But the directive also requires appropriate levels of human judgment over the use of force, and its definition of semi-autonomous systems specifically preserves operator control over the decision to select individual targets or target groups for engagement.
Anthropic's objection was narrower. It wanted its models excluded from fully autonomous weapons and mass domestic surveillance; it was not arguing that the United States should never develop or deploy autonomous weapons of any kind.
The disagreement wasn't purely technical, either. At a January appearance at SpaceX, Hegseth said the department was building “war-ready weapons and systems, not chatbots for an Ivy League faculty lounge.”
The bet that paid off
xAI made the opposite calculation.
As the Pentagon-Anthropic fight escalated in February, TechCrunch reported that xAI was preparing to become classified-ready and quoted defense-focused investor Sachin Seth of Trousdale Ventures as estimating that competitors could need six to 12 months to catch up with Anthropic's capabilities. TechCrunch also reported that, given Elon Musk's public rhetoric, xAI appeared unlikely to object to giving the Pentagon broad control over how its technology was used.
Six months later, Grok for Government is on GenAI.mil.
That doesn't prove the Anthropic dispute directly caused Grok's deployment. But the timeline makes the strategic contrast difficult to miss: Anthropic drew a line over two uses; xAI positioned itself as willing to meet the department's requirements; and the Pentagon now has Grok inside its secure enterprise AI platform.
Spreading the risk on purpose
None of this is limited to chatbots.
In May, the Pentagon announced agreements with AWS, Microsoft, Nvidia, and Reflection AI to deploy AI technologies and models on classified networks. The department described the broader strategy in terms of building an architecture that prevents vendor lock-in and gives the Joint Force access to a diverse suite of AI capabilities.
That context matters because the Pentagon's approach increasingly looks less like choosing one “official AI” and more like building a portfolio.
Gemini covers one part of the platform. ChatGPT covers another. Grok adds another. Other vendors are being wired into higher-classification environments.
The department's own language is revealing: “eliminating vendor lock” and promoting a “vibrant American AI ecosystem.”
A vendor disagreement therefore doesn't have to become an infrastructure crisis. The department can add another model.
Why this matters beyond the Pentagon
For developers watching from outside government contracting, the important signal isn't really the feature list.
It's that AI safety policies can become procurement issues when a government customer wants broader control over lawful uses than a model provider is willing to grant.
The Anthropic dispute turned that question into an actual legal and commercial fight: a company imposed usage boundaries, the government rejected them, the company was designated a supply-chain risk, and a federal court later ruled that the government's action was unlawful.
At the same time, the Pentagon has responded by broadening its AI supplier base rather than building its strategy around a single model provider.
That's the part worth watching.
The long-term question isn't whether Claude, ChatGPT, Gemini, or Grok is individually “the Pentagon's AI.”
It's whether government AI procurement is moving toward a model where no single frontier AI company is allowed to become indispensable — and where the ability to accept or reject a vendor's safety boundaries becomes part of the competitive equation.
Originally published on ZyVOP
💡 For more articles like this, subscribe to the ZyVOP newsletter!
Top comments (0)