Sen. Adam Schiff on AI regulation, free speech, and impeaching Trump one more time

In a Decoder interview with The Verge’s Nilay Patel, California Sen. Adam Schiff laid out how a Democrat who sits on the Judiciary subcommittees covering intellectual property, privacy and technology, and antitrust thinks AI should be governed. The backdrop is the non-binding pact AI CEOs signed at the White House promising to evaluate their own systems. Asked whether that pledge was toothless, Schiff agreed it was meaningless.

What he wants instead is concrete: a regulatory body with real discretion and expertise — he floated an FDA for artificial intelligence — on top of a statutory framework, because he argues Congress legislates far too slowly to keep pace. He also described bipartisan work on training-data transparency with Sen. Curtis, a data center bill he says would force operators to pay their own costs rather than pass them to consumers, and support for a narrow antitrust exemption so labs can coordinate against foreign distillation efforts.

With the midterms a month out, he said a Democratic Congress would bring vigorous oversight, including asking industry leaders to explain the times their own agents have gotten away from them. That makes this more than a podcast. It’s a preview of the hearings and bills AI companies may face next year.

Our view of the White House accord is simple: a promise to grade your own homework is not oversight. Schiff lands in the same place, and he adds a useful observation — the same executives signing a voluntary pledge in the president’s presence say publicly, elsewhere, that they want actual regulation. When an industry asks for rules and gets a pinky swear, the gap is the story.

Where I think the interview earns its time is the hard part Patel pressed on. A new AI agency sounds decisive until you remember the Supreme Court’s Loper Bright decision ended Chevron deference. Courts no longer defer to agency experts when a statute is vague, which means any well-funded lab can challenge a fast-moving AI rule and stall it for years. Schiff’s answer was to write the enabling law as specifically as possible. That’s honest, and it’s also the bind: the technology moves faster than specific statutes can, and vague statutes now invite litigation. Nobody in this conversation fully solved that, and we don’t think anyone has yet.

We’re encouraged by the narrower, workable pieces. Requiring companies to retain and disclose what copyrighted material trained their models is a sensible first step before anyone argues fair use. Making data centers carry their own grid costs answers a real local grievance. A tightly scoped antitrust exemption for security coordination — with guardrails so it can’t become a pricing cartel — mirrors the cyber information-sharing laws Schiff says he helped pass a decade ago. Those are the kinds of rules that let useful AI keep shipping while putting cost and accountability where they belong.

The Section 230 thread is the sleeper. Patel’s point is that when a platform’s own AI generates the content, it’s plausibly the company’s speech, not a user’s, and outside the old liability shield. Schiff agreed with the analysis but warned companies have the money to slow litigation. If that legal shift sticks, it will do more to change AI company behavior than any pledge — and it raises a real downside: liability fear could chill defensive tools, like security models, that are as capable on offense as defense.

The risk we’d name plainly is that AI policy gets swallowed by partisan fights — impeachment talk, merger politics, corruption charges — and the bipartisan bills Schiff describes stall the way social media reform did. He said himself that Congress did a terrible job on social media harms. If it repeats that playbook with AI, the courts and the companies end up writing the rules by default.

Schiff compared AI regulation to the long failure to rein in social media harms, arguing that the First Amendment was used as a bludgeon against rules that weren’t really speech issues, and that Section 230’s broad immunity hasn’t served its purpose. He also pointed to privacy concerns around Flock license-plate cameras and AI-driven surveillance as a bipartisan flashpoint, alongside local backlash to data centers.