AIBRIEF
Grok arrives over SpaceX's Starshield secure satellite network; ChatGPT Mil, a Pentagon tuned build, handles planning, policy, logistics, and administration.
The Pentagon's secure AI platform GenAI.mil, which is cleared to handle sensitive but unclassified work, has hit roughly 1.7 million users out of the department's more than 3 million personnel. The trigger: custom versions of Grok and ChatGPT went live the week of August 31, 2026, joining the Google Gemini version already on the platform.
Grok for Government runs over SpaceX's Starshield, the company's secure satellite network built for government entities. The Pentagon says Grok will support acquisition research, supply-chain management, and "missions faster and with greater precision" — its own framing, not an independent benchmark. ChatGPT Mil is built for planning, policy, logistics, and administration at department scale.
The Pentagon pitches the lineup as "multi-vendor, no vendor lock-in." The conspicuous gap: Anthropic, maker of the Claude model, is the only major US frontier model not on the platform. Talks reportedly broke down over "any lawful purpose" contract language, broad wording that would have permitted domestic surveillance and autonomous weapons, followed by a reported Trump-administration "supply chain risk" designation, a label typically reserved for foreign companies seen as national-security threats.
What remains unknown: whether Anthropic or any other major US frontier model maker is in active re-engagement, and whether the supply-chain-risk designation will outlast the current administration.