An AI lab calling for more regulation of itself — that sounds contradictory at first. But according to a report from WIRED, that’s exactly what Anthropic is doing at the level of US states right now. The message: existing transparency laws aren’t enough to capture the risks of increasingly capable models.
What Anthropic is actually asking for
At its core, this is about two things: independent safety audits and stronger government authority to examine systems that could threaten critical infrastructure or public safety. But Anthropic’s preferred approach doesn’t put those obligations on everyone. They’d target the largest labs — the ones above certain investment or revenue thresholds.
The company’s reasoning: building the most capable models takes enormous financial and compute resources, and it can produce capabilities that don’t exist in ordinary software. So those developers, specifically, should face additional scrutiny. Small startups stay out of scope.
The obvious objection
And it arrives right on cue. Advocates of lighter regulation argue that large incumbents could use compliance requirements to drive up costs for competitors. Whoever helps write the rules everyone has to play by gains an edge — even when the rules are well-intentioned. Anthropic counters that its proposals would apply only to companies above substantial thresholds, its own house explicitly included.
Underneath sits a bigger rift running through the industry. Some warn that a patchwork of strict state laws would fragment the US market. Others think the federal level is moving too slowly — and that states like California and New York are stepping into the gap precisely for that reason.
My take
I find this debate more honest than it looks at first glance. Yes, you can read Anthropic’s push as regulatory capture — a company pulling the ladder up and then folding part of it away. But you can also read it as what it claims to be: an attempt to set guardrails before an incident forces them.
Probably both are true at once, and that’s what makes it hard. What interests me isn’t so much who has the purer motives, but the underlying question: do we want US AI oversight shaped by individual states, by federal agencies, or effectively by the largest labs themselves? Right now it’s heading toward a mix of all three — and that gets decided not in a white paper, but law by law.
Sources: