Microsoft has developed a generative artificial intelligence model based on GPT-4 specifically for the US intelligence services. It works without an Internet connection, Ars Technica reports.

This is reportedly the first time that Microsoft has launched a language model in a secure environment to allow agencies to analyze top-secret information without the risk of leakage.

The new model, which has not yet been publicly named, is in line with the interest of intelligence agencies to use generative AI to process classified information while reducing the risk of data leakage or hacking attempts.

William Chappell, Microsoft’s CTO for Strategic Missions and Technology, said that the development of the new system included 18 months of work on modifying a supercomputer with artificial intelligence.

The modified GPT-4 model is designed to read files provided by users but does not have network access.

“This is the first time we’ve ever had an isolated version—when isolated means it’s not connected to the Internet—and it’s on a special network that’s only accessible by the US government,” Chappell told Bloomberg.

GPT-4 is a large language model (LLM) created by OpenAI. One of the serious drawbacks of using GPT-4 to analyze important data is that it has the potential to invent facts and draw false conclusions.

Since trained AI neural networks are not databases and operate on statistical probabilities, they are poor factual resources unless they are supplemented with external access to information from another source using a technique such as rich-access generation (RAG).