Schiff Pitches 'FDA for AI,' Mandatory Training-Data Retention

Senator Adam Schiff told Nilay Patel on The Verge's Decoder podcast that AI 'is crying out for legislation' and that Congress cannot repeat its yearslong drift on social media. 'It can't wait,' he said. 'We certainly can't take the kind of years we've taken to try to even wrap ourselves around social media.'

On what the regulator should look like, Schiff was willing to name the analogue. 'Whether that looks like an FDA for artificial intelligence or some other agency, I think that's ultimately going to be necessary.'

His pitch on training data was flat. 'We want to know what copyrighted material, what protected material, was used to train these models. And then we can have a discussion about what's fair use and what isn't, but we need these companies to retain that information.'

The California Democrat put self-advancing models in a national-security frame. 'As these models become more and more advanced, as they advance themselves through this recursive AI, we have less and less of a sense of what they're doing, how they're deciding what they're deciding,' he said.

Asked about the industry's voluntary pledges, per Yahoo's writeup of the interview, he dismissed them: 'I think it really is meaningless.' He also drew a tempo contrast between the labs and the Senate. 'Moore's law is increasing productivity, doubling it every 18 months, and with AI, maybe that's quadrupling it every five months. The legislative calendar works more on a Murphy's law timetable.'

The interview does not name a bill vehicle for the retention rule, and it slots into a long run of regulation coverage we've tracked this quarter.