California goes first again

Europe wrote the comprehensive law and its high-risk duties arrive in December 2027. California wrote a narrow one and it has been in force since 2 August. Scope is why.

California's AI Transparency Act became operative on 2 August, requiring watermarking, latent disclosures and detection tools from generative AI providers.

Narrow shipped, comprehensive did not

The EU AI Act needed conformity assessment, notified bodies and harmonised standards before its high-risk annex could bind. Those did not exist in time, which is why the deadline moved eighteen months.

Provenance needs none of that. You can specify watermarking without first deciding what counts as high risk — and the question California skipped is precisely the one that delayed Brussels.

The demanding clause is the detector

Watermarking is a commitment to mark. Publishing detection tools is a commitment to something falsifiable: third parties can now test whether the marking works. That is a meaningfully harder obligation than the other two and it is the one worth watching.

The firms already positioned for it made the choice before it bound. SynthID shipped on by default across YouTube-scale distribution, which is the compliance artefact this law asks for.

The best compliance posture is usually the one adopted before anyone required it.

Who is actually setting the rules

More than 100 state bills, 14 enacted, companion chatbots the most active area — and no federal statute, after Congress twice rejected a moratorium.

So the operative national regime is the union of whichever states moved first, taken at the strictest reading per question. Not a design anyone chose, and now the one everybody complies with.

Vorp Labs — US AI Regulation Update: July 2026 Laws & Policy → · TechPolicy.Press — Where State AI Legislation Stands Half Way Into 2026 → · Gunderson Dettmer — 2026 AI Laws Update: Key Regulations and Practical Guidance →