Anthropic Says Its Own AI Laws Are Already Outdated

Anthropic made headlines last year by endorsing AI transparency laws in California and New York. It was a bold move for a company that could have easily sat on the sidelines. But now the person running their US state and local policy says those laws might already be obsolete.

That’s a striking admission. You don’t often hear a company say the rules it lobbied for are already stale. But it also tells you how fast this space is moving—and how anxious Anthropic is to stay ahead of the regulatory curve.

I’ve been covering AI policy for a while now, and the pattern is always the same: lawmakers pass a bill, the industry grumbles, and then everyone pretends the law will last a decade. In reality, AI changes so quickly that any regulation is outdated by the time it’s signed. Anthropic seems to be the first major lab to admit this publicly.

Their point is not that the laws were useless. The transparency requirements in California and New York were a step forward—they forced companies to disclose more about how their models are trained and what risks they carry. That’s genuinely useful. But the underlying tech has already shifted. New model architectures, new deployment patterns, and new risks are emerging faster than the legislative process can keep up.

So what’s the alternative? Anthropic is now pushing states to regulate more quickly, not less. That sounds counterintuitive—usually companies want to slow things down. But their argument is that faster, iterative regulation is better than slow, sweeping rules that are obsolete on arrival. They’d rather have a system that updates every couple of years than a massive bill that takes five years to pass and then locks in outdated assumptions.

I think they’re right, but there’s an obvious self-interest here. If you’re a leading AI lab, you want rules that are predictable and current. You also want to shape them before they’re written. Pushing for speed gives you a seat at the table. Slower processes let other voices—and other interests—get in.

Still, I’d rather see a company advocate for faster, smarter regulation than the opposite. The real risk is that states get spooked and pass something overly restrictive in a rush. Anthropic’s challenge is to convince them that speed doesn’t mean recklessness.

Whether other AI companies follow suit remains to be seen. But if you’re watching AI policy at the state level, this is a signal that the conversation is shifting from “should we regulate?” to “how do we regulate fast enough to matter?”