The Pentagon just gave its own version of ChatGPT and Grok to every one of its 3 million military and civilian workers. That is roughly the population of a mid-sized country, all getting government-approved chatbots for writing memos, planning logistics, and digging through paperwork.
Both tools live on GenAI.mil, a secure internal portal the Pentagon built last year so staff would stop pasting sensitive documents into consumer chatbots. Gemini from Google was the first model on the platform. Now ChatGPT Mil and Grok for Government have joined it, and more than 1.7 million people out of the 3 million eligible have already signed up.
The more interesting story is who is missing. Anthropic's Claude is nowhere on this platform, and that absence did not happen by accident.
Back in 2025, the Pentagon asked four AI companies, OpenAI, Google, xAI, and Anthropic, to strip away the safety limits on their chatbots so the military could use them without restriction. Three companies agreed. Anthropic refused two specific requests: letting its AI make fully autonomous decisions in weapons systems, and using it for mass surveillance of American citizens.
When talks fell apart, Defense Secretary Pete Hegseth labeled Anthropic a national security supply chain risk in March, a status normally reserved for foreign threats. That label forced every Pentagon contractor to certify they were not using Claude at all, cutting Anthropic out of a huge chunk of government business almost overnight.
Anthropic sued. On August 27, a federal judge ruled the whole thing was illegal, calling the blacklisting "illegal and baseless" and finding it amounted to punishing the company for pushing back on policy. Four days later, the Pentagon launched ChatGPT and Grok anyway, with Claude still absent from the lineup.
For any business that buys software from a handful of big vendors, this is worth watching closely. It shows that a powerful customer, in this case the largest employer in the country, can pressure an AI company to drop its own safety rules, and a company that says no can lose access even when a court later says the punishment was unlawful. Legal victories do not automatically restore business relationships.
It also shows why the Pentagon runs three or four AI models side by side instead of picking one. If a vendor becomes unavailable, for legal, political, or business reasons, the work does not stop. Any company depending heavily on a single AI tool for daily operations should treat that as a warning, not a Pentagon-specific story.