WASHINGTON — The Pentagon announced Monday that OpenAI’s ChatGPT will join xAI’s Grok and Google Gemini on its GenAI.mil platform — with a Pentagon official telling Breaking Defense this morning that the site might “potentially” add a fourth model in the future.
Sec. Pete Hegseth and Pentagon CTO Emil Michael officially launched the site just two months ago to bring popular Large Language Models to all 3 million uniformed military, civil service, and contractor employees of the Defense Department. Posters bearing Hegseth’s likeness over the exhortation “I want you to use to use AI: Go to GenAI.mil today” have been plastered prominently on Pentagon walls.
The site now claims 1.1 million unique users and rising, with no downtime or other major technical glitches since launch. (That said, an official DoD news story acknowledges the site’s surprise rollout made many Pentagon employees initially suspect it was a hoax, a virus, or a cybersecurity test). All three military service departments within the Department of Defense — the Departments of the Army, the Navy (which includes the Marines), and the Air Force (which includes the Space Force) — have announced their adoption of GenAI.mil, with the Air Force’s post on xAI specifically saying it is “phasing out legacy AI systems” from early attempts to adopt commercial models.
Having a wide range of AI models was the plan for the platform from the beginning. GenAI launched with just Google’s Gemini, then added the controversial Grok during the quiet period three days before Christmas, and is now deploying ChatGPT — the OG generative AI that kicked off the current AI revolution just three years ago. All three AIs on the site are modified versions of the publicly available models, approved by the Pentagon for use on sensitive but unclassified data. Clearance for use on “secret” and “top secret” data is a much higher hurdle.
OpenAI describes what’s going on GenAI.mil as “a custom version of ChatGPT [that] runs in authorized government cloud infrastructure with built-in safety controls and protections for the Department’s highly sensitive data,” according to the company’s press release Monday. “Data processed on GenAI.mil remains isolated to the government environment and is not used to train or improve OpenAI’s public or commercial models, ensuring separation from external systems and protecting mission data.” (The tendency of commercial AIs to store user prompts and add them to their training data was a major concern for Pentagon IT and cybersecurity officials early on.)
This version of ChatGPT derives specifically from OpenAI for Government, the company’s push to customize its Large Language Model for federal users, which began in June with a contract from the Pentagon’s Chief Digital & AI Officer (CDAO) worth up to $200 million. Weeks later, CDAO followed up with similar $200 million contracts for xAI, Google, and Anthropic.
The Pentagon official who spoke to Breaking Defense was cautious about what might come next. While they declined to say whether or not adding a fourth model was “probable,” they did say it might happen “potentially in the future.”
It’s worth noting that, of these four prominent “frontier models” that won awards last summer, only Anthropic’s Claude is not yet available through GenAI.mil. That suggests that, if GenAI.mil continues its expansion, Claude — which in November suffered a major subversion of its cybersecurity tools by Chinese hackers — may be the next likely candidate.
