Meet the Startup Making GPUs More Secure for AI

When I ask business executives about their use of cloud providers’ generative AI tools, I’m often reminded that many firms don’t yet process sensitive data in cloud servers because they’re concerned about the security of the data. Cloud providers have long used encryption to protect data when it’s stored or moving from one location to the next. But it’s much harder to secure data while it is in use, say when a person analyzes a spreadsheet or messages an AI chatbot. 

Various efforts have been tried to fix this problem. Some technology companies have adopted a technique known as confidential computing. This involves processing data in a secure enclave—essentially a protected black box—of a computer’s central processing unit. Apple, for instance, developed a version of this method for processing data from AI applications.

But EQTY Lab, a Switzerland-based startup that spun out of Stanford University, believes confidential computing does not go far enough, especially when it comes to protecting the security of AI applications. The company has been working with Nvidia and Intel on a new effort the companies call verifiable computing, which they’ll unveil later Wednesday.

EQTY’s software takes advantage of the current security features in Nvidia’s graphics processing units (the H100 or H200) and Intel processors. EQTY serves as a digital notary embedded in the chip that produces certificates about what went on during a computing process.