The Pentagon’s wager is that AI can strip friction from military paperwork and planning without loosening control over sensitive data. Its answer to both productivity demands and security fears is a curated, multi-model system rather than reliance on a single commercial provider.

On Aug. 31, the Department of War added Starshield AI’s Grok for Government and OpenAI’s ChatGPT Mil to its GenAI.mil platform, extending an effort launched nine months earlier. More than 1.7 million of the department’s over 3 million personnel have since been onboarded, according to the department.

The two systems arrive after separate partnerships with OpenAI for deployment on classified networks and with Elon Musk’s xAI to assist work involving controlled unclassified information. Both new tools are accredited at impact level five, meaning they are intended for unclassified material that nevertheless requires safeguards.

For the Pentagon, the appeal is practical. Grok is being pitched for tasks from supply-chain management to market research, with customizable workspaces, persistent projects and reusable playbooks. Officials said it could deliver “immediate productivity gains, stronger knowledge continuity, and more secure and efficient collaboration.”¹

ChatGPT Mil, meanwhile, is designed around chat, files, projects and custom GPTs. The department says it will handle document-heavy work in planning, logistics, administration and policy—routine burdens it hopes to move faster so personnel can focus on higher-priority Joint Force work.

The shared strategy is not simply to add fashionable tools. By placing ChatGPT and Grok alongside existing systems, the Pentagon says it is building a resilient AI ecosystem and reducing dependence on any one vendor. That is also a security argument: in June, the Defense Counterintelligence and Security Agency warned that “shadow AI”—unapproved outside models or unevaluated embedded features—creates an unmonitored attack surface where risks can outrun safeguards. The Pentagon’s message is that military AI use is coming either way; the contest is whether it happens within approved guardrails.