Microsoft has created a generative AI model for top-secret work within US intelligence agencies.
This model is supported in an « air-gapped » environment, Microsoft’s CTO of Strategic Missions and Technology William Chappell told Bloomberg. In other words, it’s entirely unconnected, distinct from cloud-based AI models that are susceptible to hacks and data leaks.
What makes Microsoft’s new AI model a coveted spy gadget
For intelligence agencies, Microsoft’s model is a way to process and analyze vast quantities of classified information without compromising secrecy. Last fall, Bloomberg reported that the CIA was building its own internal ChatGPT-style tool to sift through public data for clues. In April, Sheetal Patel, assistant director of the CIA for the Transnational and Technology Mission Center, said intelligence agencies all over the world are racing to be the first to harness generative AI, saying « I want it to be us. »
The model, built upon OpenAI’s GPT-4 large language model, underwent an 18-month development process, according to Chappell. The project required overhauling a supercomputer in Iowa and is on a special network that’s only accessible by the US government. While the model is technically cloud-based, it’s a static cloud that’s isolated from the internet, so it can’t learn from data and files shared by users or information gleaned from the web.
Chappell said the model is now live and undergoing testing by the various US intelligence communities.
