Pentagon Adds ChatGPT and Grok to Military AI Portal, Leaves Out Anthropic

The Pentagon opened its enterprise artificial intelligence portal to OpenAI's ChatGPT and xAI's Grok on Monday, putting three rival chatbots — and, notably, not Anthropic's Claude — inside the secure network used by the Defense Department's roughly 3 million military and civilian personnel.

The additions to GenAI.mil, a centralized portal the department launched last year, bring the platform's roster to three frontier chatbots: Google's Gemini for Government, which has been available since launch, a defense-tailored version of ChatGPT called ChatGPT Mil, and Grok for Government, offered through Elon Musk's xAI under its Starshield government arm. All three models have been authorized to handle Impact Level 5 data, the Pentagon's classification tier for sensitive but unclassified information such as personnel records and operational planning documents.

GenAI.mil has already onboarded 1.7 million unique users out of the department's roughly 3 million-person workforce, according to figures reported by DefenseScoop. ChatGPT Mil arrives with chat, file upload, project workspaces and custom GPTs, and is pitched at document-heavy work such as travel forms, contracting research and policy drafting; a Pentagon official said access to GPT-5.6 Terra is expected soon, with the current deployment running on GPT-5.4 Terra. Grok for Government adds adaptive reasoning modes the company calls Auto, Fast and Expert, along with persistent project workspaces and reusable prompt "playbooks."

A deliberately crowded field

The Pentagon has framed the multi-vendor approach as a hedge against dependence on any single AI company. "Access to a diverse suite of AI capabilities from across the resilient American technology stack will give warfighters the tools they need to act with confidence," a defense official said, according to DefenseScoop's reporting.

The rollout follows separate infrastructure agreements the department has struck with Amazon Web Services, Microsoft and Nvidia to extend AI compute onto classified networks, part of a broader push to normalize commercial AI tools inside a bureaucracy that has historically been slow to adopt them. GenAI.mil itself was built as a walled garden specifically so that department staff could query commercial frontier models without routing sensitive prompts or documents through the same consumer-facing servers that handle ordinary ChatGPT or Gemini traffic — a distinction officials have repeatedly stressed as they try to convince a wary workforce that using AI on the job will not leak government data.

The department has also cast the expansion in competitive terms, noting that giving personnel a choice between models is meant to avoid the kind of single-vendor lock-in that has drawn criticism in other federal IT contracts. A Defense One report cited by DefenseScoop quoted an official saying Gemini has proven useful for search-style queries while ChatGPT tends to get used more for long-form text work, suggesting the department expects staff to mix and match tools by task rather than standardize on one.

Absent from Monday's expansion is Anthropic, whose Claude models remain excluded from GenAI.mil. TechCrunch reported that the standoff stems from a contractual dispute in which the department sought "unrestricted access" to Claude for defense applications, while Anthropic held firm on safeguards it maintains against uses such as mass surveillance and lethal autonomous weapons systems. The administration has since designated Anthropic a "supply-chain risk," a label the company is contesting in court; a federal judge already ruled against that designation on Aug. 28, and a fuller account of the ruling that found Hegseth's designation was unlawful retaliation lays out how the court reached that conclusion.

The Anthropic exclusion has drawn attention precisely because it cuts against the department's stated preference for choice. Anthropic has built its brand around stricter usage policies than rivals, including commitments not to support fully autonomous weapons targeting or bulk domestic surveillance, and has said it will not loosen those terms even for government contracts of this size. The "supply-chain risk" designation the administration has since applied to the company is an unusual step against a major U.S. AI lab, and Anthropic's legal challenge to it is being watched closely by other vendors weighing how much leverage the Pentagon actually has in these negotiations.

What's next

Defense officials have described GenAI.mil as a living platform rather than a finished product, with additional model upgrades and expanded classification coverage expected in the coming months. The department has signaled that a jump from GPT-5.4 Terra to GPT-5.6 Terra is imminent for ChatGPT Mil, and officials have suggested that offline, citation-backed search capability will keep expanding as more document repositories are indexed for the system.

The dispute with Anthropic is likely to remain a flashpoint as the department weighs how much latitude to give AI vendors over military use cases, at a moment when the Pentagon's AI budget and vendor relationships are drawing closer scrutiny from Congress; lawmakers on defense and technology oversight committees have asked for briefings on how GenAI.mil vets which models get access to sensitive data, and what recourse exists for vendors who decline the department's terms. For now, the three-way split between OpenAI, xAI and Google sets up a live comparison of how differently postured AI labs perform under the specific constraints of government and defense work — and a test case for whether a model's usage policies, rather than its raw capability, become the deciding factor in which companies get to build the U.S. military's AI infrastructure.