By David Shepardson
WASHINGTON (Reuters) -The Biden administration will soon propose requirements for cloud companies aimed at finding out who is accessing U.S. clouds to train AI models, U.S. Commerce Secretary Gina Raimondo said on Friday, as the burgeoning AI sector raises increasing security concerns.
‘We can’t have non-state actors or China or folks who we don’t want accessing our cloud to train their models,” Raimondo said in an interview with Reuters.
“We use export controls on chips,” she noted, adding, “Those chips are in American cloud data centers so we also have to think about closing down that avenue for potential malicious activity.”
The proposed “know your customer” regulation is set to be released as soon as next week. “It is a big deal,” Raimondo said.
The Commerce secretary said U.S. cloud computing companies “should have the burden of knowing who their biggest customers are training the biggest models, and we’re trying to get that information. What will we do with that information? It depends on what we find.”
Raimondo spoke to Reuters after speaking at an event with former Secretary of State Condoleezza Rice, telling an audience the department was “beginning the process of requiring U.S. cloud companies to tell us every time a non-U.S. entity uses their cloud to train a large language model.”
President Joe Biden in October signed an executive order requiring developers of AI systems that pose risks to U.S. national security, the economy, public health or safety to share the results of safety tests with the U.S. government, using the government’s authority under the Defense Production Act (DPA), before they are released to the public.
Those survey requests under the DPA are soon going to companies, Raimondo said, telling Reuters that companies will have 30 days to respond. “Any company that doesn’t want to comply is a red flag for me,” she said.
Top cloud providers include Amazon.com’s AWS , Alphabet’s Google Cloud and Microsoft’s Azure unit.
(Reporting by David Shepardson; Editing by Leslie Adler)
