May 8, 2024 Microsoft Corp. has launched a generative AI model that operates entirely without internet access, tailored specifically for US intelligence agencies. This marks the first deployment of a major language model fully isolated from the web, addressing the critical need for top-level security in handling sensitive information.

The AI, based on GPT4, is designed to assist intelligence operatives in managing and analyzing the vast volumes of classified data generated daily. Unlike typical AI models that depend on cloud services and internet connectivity to process and learn from data, Microsoft’s AI operates in a completely air-gapped environment, ensuring that no data can leak or be exposed to cyber threats.

William Chappell, Microsoft’s Chief Technology Officer for Strategic Missions and Technology, highlighted the significant engineering challenges overcome during the 18-month development period. The project involved retrofitting an AI supercomputer in Iowa to function securely within a government-exclusive network.

This AI deployment comes as intelligence agencies globally express a growing interest in utilizing generative AI for data analysis without compromising data security. The CIA, for instance, has already introduced a ChatGPT-like tool at unclassified levels but required a more secure solution for handling classified data.

Microsoft’s initiative not only strengthens national security but also positions the US at the forefront of utilizing AI technology within the intelligence community. The model is now live, providing functionalities like answering queries and writing code, and is set for further testing and accreditation by the intelligence community.

This strategic development underscores the potential of AI in national security while addressing the imperative of maintaining stringent data confidentiality.