Equities
Microsoft deploys offline, air-gapped generative AI for US intelligence, ensuring top-secret data analysis with GPT4 model security.
By Athena Xu
ᐧ
Microsoft Corp. has achieved a significant milestone by deploying a generative AI model that operates entirely offline, marking a first in the realm of large language models. This development caters specifically to the needs of US intelligence agencies, enabling them to analyze top-secret information with the advanced capabilities of generative AI while ensuring the utmost security. William Chappell, Microsoft’s chief technology officer for strategic missions and technology, highlighted the uniqueness of this deployment, emphasizing its complete separation from the internet to maintain a secure environment.
The AI model, based on GPT4, and its supporting infrastructure have been installed on a cloud platform that is "air-gapped," meaning it is isolated from the internet. This setup addresses the intelligence community's demand for generative AI tools that can process classified information without the risk of data leakage or hacking. The initiative reflects Microsoft's commitment to delivering a truly secure system, as explained by Chappell, who has a background in developing microsystems for the Defense Advanced Research Projects Agency (DARPA).
The deployment of this AI model is a response to the intelligence community's desire to leverage generative AI for analyzing the vast amounts of classified data produced daily. Sheetal Patel, assistant director of the CIA for the Transnational and Technology Mission Center, expressed the strategic importance of being the first to utilize generative AI for intelligence purposes, emphasizing the competitive advantage it would provide. Microsoft's efforts over the past 18 months, including the overhaul of an AI supercomputer in Iowa, underscore the company's dedication to this goal.
The GPT4 model implemented in this secure cloud is static, meaning it can process files but is incapable of learning from them or any external internet sources. This design choice ensures the model remains "clean," safeguarding sensitive information from being inadvertently incorporated into the platform. Chappell mentioned that approximately 10,000 individuals could potentially access the AI, highlighting the balance between accessibility and security. The service, which became operational recently, is now subject to testing and accreditation by the intelligence community to ensure it meets all necessary standards and requirements.
"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." "You don’t want it to learn on the questions that you’re asking and then somehow reveal that information." "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."
Finance GPT
beta