Microsoft It has been renewing many of its products with artificial intelligence for just over a year. Copilotwhich uses the language model GPT-4 developed by OpenAI that is behind ChatGPT Plus, has been integrated with Windows 11, although not in Spain, Microsoft 365, the Bing search engine and other services intended for companies and/or the end consumer. Now we have known that US intelligence agencies enter this group after the launch of a new AI chatbot from Microsoftalso based on GPT-4.
This new artificial intelligence service, which for the moment does not have a public namehas been created to be used by organizations such as the INC and it has been available since last week, although its existence was revealed this Tuesday by Bloomberg. The AI assistant has been designed to allow spy agencies to analyze top secret information without connectivity risks and to enable secure conversations with a chatbot that works similarly to how ChatGPT and Copilot do.
According to the media, the new AI service responds to a growing interest among intelligence agencies in use generative AI to process classified data, while mitigating the risks of data breaches or hacking attempts. ChatGPT runs on cloud servers provided by Microsoft, which may present risks of interception and data leakage. The CIA announced its plan to create a service similar to ChatGPT last year, but this Microsoft release is a separate project.
The development of the new system involved 18 months of work to modify an artificial intelligence supercomputer in Iowa. The modified GPT-4 model is designed to read files provided by its users, but cannot access the open Internet. “This is the first time we have an isolated version, when isolated means that it is not connected to the Internet, and It is on a special network that only the United States government can access“he pointed out William ChappellMicrosoft's chief technology officer for strategic missions and technology, to Bloomberg.
While Microsoft's new spy AI service is now available, must still be tested and approved by US intelligence agencies. It is not known how long this trial period will last or what price Microsoft plans to charge the government for this service. artificial intelligence spy, but some 10,000 people in the intelligence community can now access it. He is currently “answering questions,” according to Chappell.
In March, the US House of Representatives Office of Cybersecurity banned the use of Microsoft's Copilot generative AI assistant by its staff. The US House of Representatives said there was a “Risk to users due to threat of leaking Camera data to unapproved cloud services.”
By operating within a network isolated from the Internet, this problem should be solved. Another question is how a language model like GPT-4 works, very simply predicting the probability that one word follows another in its responses, and his tendency to fable and invent things. GPT-4 may provide inaccurate summaries, draw erroneous conclusions, or give false information. That is, in this case, misinform intelligence agencies. The supervision and safeguards that the system will have to prevent this are unknown.