Pentagon adds ChatGPT and Grok to its staff AI platform, alongside Google Gemini

GenAI.mil, the Pentagon's secured AI platform, now hosts three commercial vendors at once. Google Gemini was already on it; OpenAI's ChatGPT Mil and xAI's Grok for Government joined this week, the latter labeled in the Defense Department's press release as a "Starshield AI" build. About 1.7 million of the department's 3 million civilian and military staff are already active on the platform, and the new models are joining an installed base, not standing one up.

The same secured runtime now hosts Gemini, ChatGPT Mil, and Grok for Government, each in a defense-grade configuration rather than the consumer product. The Pentagon frames both new models as tailored to "warfighter needs." For the secured Grok variant, the press release cites "immediate productivity gains, stronger knowledge continuity and more secure and efficient collaboration," with use cases running from market research for acquisition staff to supply-chain work for logisticians and "faster, more precise mission execution across numerous operational contexts." ChatGPT Mil is positioned as a day-to-day assistant for document-heavy work and file-integrated chats, the kind of task profile a 3-million-staff workforce generates without needing a frontline mission.

The press release does not publish the deployment timeline, the contract vehicles, or the security clearance levels that govern each build. It does not define what "secured" buys in practice: a Department of Defense-operated runtime, isolated data, role-based access, or a thinner certification pass. For a procurement this large, those are load-bearing details, and the rollout has been announced without them.

Anthropic was designated a supply-chain risk by the Pentagon after the company refused to allow Claude to be used for autonomous weapons and mass domestic surveillance. That designation was recently overturned, putting Anthropic back in the vendor pool just as xAI enters it. OpenAI, Google, and xAI each shipped a DoD-ready build. Anthropic fought the framing, lost, won on appeal, and is now in the same pool. The reversal reads cleanly: a willingness to integrate with defense procurement, and a posture the Pentagon can live with, have mattered more than any model benchmark.

Two questions remain open: what "secured" actually buys at the architecture level, and whether each build on GenAI.mil meets the same bar; and whether the Anthropic reversal was the end of a procurement fight or the start of a new one, since the supply-chain-risk mechanism that once blocked Claude can be reapplied to any vendor that later draws the same line. The Pentagon's "warfighter needs" framing is broad enough to cover both routine document work and live operational decisions, and the policy has not yet drawn that line publicly.

The department says more models are on the way. The supply-chain-risk mechanism that hit Anthropic is still on the books, and any future vendor that draws the same line will face the same review.