© Mikhail Nilov Pexels.com
A few hours ago, Microsoft launched a generative AI model dedicated to American intelligence agencies. The goal ? Analyze ultra-confidential information.
This model is based on GPT-4, but its greatest particularity is that it is designed to operate entirely disconnected from the Internet, which allows maximum security of the processed, and often classified, data.
A disconnected ChatGPT, designed for spies
During the presentation of this new Intelligence offer Artificially, the CTO in charge of strategic missions at Microsoft, William Chappell, said “This is the first time we have had an isolated version – where isolated means that’ ;it is not connected to the Internet – and it is on a special network only accessible by the American government, ”.
Even though this version of GPT-4 is based in a cloud infrastructure, it uses an “Air-Gapping” which allows it to be entirely separated from the internet.
Another crucial point, this version of GPT-4 will not use files and documents transmitted to improve learning, and therefore allows each defense secret information to be compartmentalized.
This is the first time that a communication ( official) presents an AI model built specifically for classified workloads. Last year, the CIA deployed an alternative to ChatGPT to filter massive amounts of public information. However, this AI model has not been used on classified documents. Other government agencies are using generative AI to work with unclassified data.
This Artificial Intelligence for spies is the result of 18 months of work for Microsoft, which used a supercomputer dedicated to Artificial Intelligence in Iowa. Even if the model has been operational for more than a week, it must still undergo a battery of tests, and must above all receive validation from the intelligence services to be used on a daily basis.
📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.
[ ]