Senator Schiff wants a real AI agency, not another White House photo op

The gap between what the White House is doing on AI and what Adam Schiff thinks needs to happen is pretty wide. When tech CEOs gathered at the White House to sign a nonbinding pledge about evaluating their own AI systems, Schiff saw it as exactly what it was — a promise with no teeth. His position: you need an actual federal agency with rulemaking authority, not a handshake.
Schiff's argument isn't just about AI being risky. It's about how you enforce anything in the current legal environment. He points to the Supreme Court's decision in Loper Bright overturning Chevron deference. For decades, courts deferred to expert agencies when laws were vague. Now they don't. So any rule an agency writes — say, a new AI safety standard — immediately gets challenged in court, and judges who aren't AI experts get to decide what the law means. That's a recipe for years of litigation before anything actually takes effect.
That's the core reason he supports creating a dedicated AI agency. Not because he thinks AI is uniquely scary, but because the existing patchwork of regulators can't move fast enough, and the courts have made it worse. A new agency would have explicit statutory authority to write rules, and Congress would have to be specific in the law itself so courts can't just wave it away.
He also connects this to the broader political moment. Schiff sees the current administration as chaotic and distracted — campaign ads funded by taxpayer dollars, press locked out of briefings, and the bizarre claim that the triumphal arch being built in DC is a national security project because they want to store ammunition inside it. The arch bit is genuinely funny but also revealing: the administration reaches for national security justifications for everything, but didn't bother using that argument to make the AI pledge binding.
If Democrats retake Congress in the midterms, Schiff says AI regulation becomes a real priority. But the window matters. Every year you wait, the models get more capable and the precedent gets weaker. He's not calling for a pause or a moratorium — just a structure that can actually enforce whatever rules get written.
One thing I found interesting in the conversation: he didn't shy away from comparing AI regulation to social media. The lesson there is that letting platforms self-regulate worked for a while, then stopped. AI is on the same trajectory — voluntary commitments are fine until someone actually gets hurt or a model does something embarrassing in public.
On impeachment: Schiff didn't rule out supporting a fourth impeachment of Trump over the corruption stuff. He framed it as a question of whether the evidence rises to that level, which is more restrained than I expected given how he opened the interview.
The practical takeaway for anyone building with AI: don't wait for regulation to figure out safety. Schiff's own view is that the White House pledge is insufficient, and any real rules are at least a midterm election away — and then years of court fights after that. Your customers won't wait, so your own evaluation and testing process is the only safety net you'll have for a while.