Senator Schiff: AI rules are urgent, and Congress has a poor track record from social media

Adam Schiff is a Democratic senator from California and sits on several committees dealing with tech and AI: copyright, competition and privacy. In an interview on the podcast Decoder from The Verge, he commented on a voluntary agreement in which top AI executives at the White House promised to keep AI safe.

His verdict is clear: the agreement is meaningless. It is not binding, and it merely requires the companies themselves to make sure their controls and monitoring work. Schiff points out that several of the same executives have publicly called for actual regulation.

He wants a separate agency for AI, perhaps along the lines of the FDA for medicine, with real power and professional expertise. The reason is the pace: Congress is too slow to keep up with the technology. There is a catch, though. The Supreme Court has abolished the principle that courts defer to agencies' experts when a law is imprecise. New rules can therefore quickly end up in court. Schiff believes the laws must be written as precisely as possible.

On copyright, he is working with a Republican senator on a requirement that AI companies must keep records of which protected material they have trained the models on. Only then can a position be taken on what constitutes fair use. He also points out that privacy is under pressure, because powerful AI can be used to monitor people and collect data about them.

Schiff believes that AI skepticism cuts across party lines, including in opposition to data centers and fear of lost jobs. His own bill on data centers aims to ensure that they pay their own costs, so consumers do not. If the Democrats win a majority in the midterm elections, he expects thorough oversight of the industry and attempts at legislation.

What does it mean for you? This is American politics, and nothing new has been passed that changes your everyday life with AI, either privately or at your company. You do not need to act now. A sensible rule of thumb is when concrete, enacted rules arrive on, for example, training data or liability. Only then can it affect the tools and terms you use.