Microsoft Strategic Missions and Technology CTO William Chappell announced that it’s deployed a GPT-4 large language model in an isolated, air-gapped environment on a government-only network. Bloomberg first reported the setup, citing an unnamed executive who claimed that the Azure Government Top Secret cloud-hosted model represents the first time a “major” LLM has operated separated from the internet.

Chappell announced the AI supercomputer on Tuesday afternoon at the “first-ever AI Expo for National Competitiveness” in Washington D.C. Unlike the models behind ChatGPT or other tools, Microsoft says this server is “static,” operating without learning from the files it processes or the wider internet.

Chappell told Bloomberg, “It is now deployed, it’s live, it’s answering questions, it will write code as an example of the type of thing it’ll do.” As Chappell mentioned to DefenseScoop, it has not been accredited for top-secret use, so the Pentagon and other government departments aren’t actually using it yet, whether that’s processing data for a particular mission or something like HR.