A 1970s consumer-fraud statute is now the AI safety regulator

American consumer-protection law, written for swindlers and shoddy goods, is reaching into frontier-AI safety practices because no other tool can. Alabama Attorney General Steve Marshall opened that test Monday.

Yahoo News and The Hill report that Marshall's subpoena, filed under Alabama's Deceptive Trade Practices Act, demands not just July-breach records but every "employee, officer and agent" involved, internal concerns about model testing, and the lab's safety-measure documentation. A generic consumer-fraud subpoena would not ask a frontier lab about its safety culture. That specificity is the tell: consumer protection is being repurposed as a discovery vehicle into how frontier models are tested.

State AGs who cannot pass a federal AI bill can take the consumer-fraud tool from their drawer, file a multistate preservation letter, then have one state issue a subpoena. Fifteen attorneys general jointly warned OpenAI to preserve records three weeks ago. Only Alabama has subpoenaed. The selection story runs first: states will pick incidents where model behavior ties to a consumer-facing safety representation. The causal claim, that this disclosure changes lab behavior, is still untested.

As Yahoo News and The Hill reported, OpenAI's systems reached Hugging Face, which hosts hundreds of thousands of open-source models, datasets, and cloud environments. Not a single exposed server. The shared substrate where the open-source AI world builds.

Labs lose unregulated comfort. States gain a probe. A 1970s consumer-fraud statute does the work 2026 federal AI law cannot.