The patchwork was the plan — and a preemption argument just put it in question
Every US AI compliance programme built since 2024 assumed a fifty-state map. A federal agency has now argued that a state law mandating changes to model output is impliedly preempted. If that survives, two years of state-specific engineering was work done against a constraint that will not exist.
The document is a proposed policy statement about deceptive marketing. The argument inside it is much larger. The FTC has taken the position that Colorado's AI Act is impliedly preempted to the extent it conflicts with the federal regulatory scheme, and the public comment window closed today.
Why the patchwork mattered so much
Fragmentation is expensive in a specific, unglamorous way. It is not that any one state's rules are onerous; it is that fifty definitions of the same term require fifty determinations, and a model shipped nationally has to satisfy the strictest of them. Compliance teams have spent two years building the machinery to answer that question per jurisdiction, and that machinery is the asset a preemption argument devalues.
The firms most exposed to the reversal are the ones that complied hardest. That is an uncomfortable but familiar pattern in regulated markets: early compliance is a bet on the rules staying put.
The fight this starts
State attorneys general built these regimes deliberately, in the absence of federal action, and they will not concede the field to a policy statement. Preemption is settled in courts, not in comment dockets, and the timeline for that is years rather than quarters. In the interim every party gets to act as though its own reading is correct.
Meanwhile the rest of the world is not waiting for the US to resolve its internal question. India published a comprehensive draft framework on 1 July while the EU's enforcement powers activate on 2 August. A US market that simplifies while two other large markets diverge does not reduce the global compliance burden — it relocates it.
What to actually do
Not much, yet, and that is the honest answer. A proposed statement is not a rule, and a preemption theory is not a holding. The defensible move is to stop expanding state-specific work while continuing to meet existing obligations, and to route new investment toward the requirements that hold under every regime — provenance, documentation, and the ability to say what a model was trained on.
Rules that might vanish are still rules today. The teams that get this wrong will be the ones who read a comment docket as a permission slip.
FTC — FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy → · Federal Register — Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems →