Microsoft’s top-secret generative AI for US spies: What you need to know
Microsoft launched its most recent generative AI model on Thursday, according to a Bloomberg report. The model was made specifically for “U.S. intelligence agencies.”
This large language model is “entirely divorced from the Internet,” a first for the industry that will allow intelligence agencies to analyze classified information without fear of it accidentally leaking to the Internet or being hacked.
We’ve seen how generative AI has improved major businesses, from automating tasks to generating new ideas to bounce off of and streamlining data. And understandably, intelligence officials want these generative AI tools to help with their insane amounts of data.
The CIA has already been experimenting with a “ChatGPT-like service at unclassified levels,” but intelligence agencies also have a ton of classified information to sift through, and a large language model isolated from the Internet is the perfect solution.
Much like the space race, countries are competing to “get generative AI onto intelligence data,” Sheetal Patel, Assistant Director of the CIA for the Transnational and Technology Mission Center, told Bloomberg. The first country to do so would be the winner, and Patel said, “I want it to be us.”
Microsoft wants to be the company to deliver this historical generative AI service to the government, at least in part to redeem itself from trailing behind Google’s AI advancements.
When did Microsoft start developing AI for the US government?
It may seem a bit out-of-the-blue for Microsoft to launch a generative AI model for the government, but the company’s been working in this field for at least a few years. With this recently deployed LLM specifically, Microsoft spent the last 18 months working on it.
Microsoft is obviously working on improving Copilot for more general AI experiences on PCs, but the company is also invested in how AI can support government services and the public sector.
Microsoft says that AI can be used to “streamline systems, prevent fraud, and improve citizen engagement” within government services and talks about the possibility of “AI-powered smart cities,” which sounds like it’s straight out of a sci-fi novel.
When discussing AI in the public sector, Microsoft points out how generative AI can act as a chatbot for citizen services, a creative aid for first drafts of projects and proposals, and a data-entry wizard for info that “may have been previously analyzed separately and manually.” That last point is especially applicable to intelligence agencies.
We might not have heard about Microsoft’s involvement in government-related generative AI 18 months ago. Still, in October 2023, Candace Ling wrote in a Microsoft blog post that it “began enabling government agencies to adopt new AI technologies for mission-critical solutions with Azure OpenAI Service in the Azure commercial cloud.”
Ling went on to write that Microsoft “[recognizes] that many agencies handle sensitive data requiring higher levels of security and compliance” and that it would be rolling out applicable solutions over the next few months.
And now, we have the first look at that solution: a large language model that’s “on a special network that’s only accessible by the US government,” according to William Chappell, Microsoft’s CTO for Strategic Missions and Technology. Chappell told Bloomberg that the model is static, “meaning it can read files but not learn from them, or from the open internet,” which will prevent secret, classified info from getting absorbed into it.
The model officially went live on Thursday, but it “now needs to undergo testing and accreditation by the intelligence community” before it can start widespread use among intelligence agencies.
