Microsoft develops internet devoid generative AI service for US intelligence agencies
Microsoft (NASDAQ:MSFT) has developed a generative AI model entirely devoid of the internet for the U.S. intelligence agencies, so that the technology can be safely used to evaluate top-secret information, Bloomberg News reported.
It is the first time a major large language model, or LLM, has operated fully divorced from the internet, said a senior executive at the company, as per the report.
Most AI models, including ChatGPT from OpenAI — in which Microsoft has invested billions of dollars — rely on cloud services to learn and infer patterns from data. However, the company wanted to provide a truly secure system to the U.S. intelligence agencies.
Microsoft has deployed the GPT4-based model and vital elements which support it onto a cloud with an “air-gapped” environment that is isolated from the internet, said William Chappell, Microsoft’s chief technology officer for strategic missions and technology.
The U.S. tech giant spent the last 18 months working on its system, including overhauling an existing AI supercomputer in Iowa.
The GPT4 model placed in the cloud is static, which means that the AI model can read files but not learn from them or from the open internet. Theoretically, about 10,000 people would be able to access the AI, according to Chappell.
The service had gone live last Thursday and would need to be subjected to testing and accreditation by the intelligence agencies, the report added.