Pick a side. That’s essentially what the current U.S. AI regulatory environment is asking AI developers to do.
Not explicitly, not yet. But the gap between federal intent and state action has never been more visible than it was on March 26, 2026, the day Washington State Governor Bob Ferguson signed two AI bills into law and the day the reporting on the White House’s National Policy Framework for Artificial Intelligence was reaching compliance desks across the country. The framework, released March 20, asks Congress to preempt state AI laws. The laws Washington State just enacted are exactly the kind the framework has in mind.
What Happened, and Why the Timing Is Not Incidental
The White House framework is a comprehensive set of more than two dozen legislative recommendations built on a December 2025 executive order. Its central structural recommendation asks Congress to establish a single federal AI standard and preempt state AI laws that “impose undue burdens” on AI development. Legal analysts have described it as the administration’s most complete statement yet of what federal AI law should look like.
On the same day the framework’s implications were still being absorbed, Governor Ferguson signed two laws. The first, HB 2225, establishes chatbot safety requirements, including provisions that, according to initial reporting, address topics such as suicide prevention for minors. The second requires AI developers meeting a covered-entity size threshold to provide tools enabling detection of AI-generated video, audio, and images. Both signings were confirmed by KNKX Public Radio and corroborated by independent outlets including KUOW and MLex.
States aren’t waiting. That fact, not the specific contents of either Washington law, is the operative reality the framework is trying to address.
The Legal Gap Between a Recommendation and a Law
The framework is not a law. This is not a nuanced legal point. It is the most important sentence in any compliance briefing about this document.
Federal preemption of state law happens when Congress acts, through express statutory language overriding state rules, or through a regulatory scheme so comprehensive that state law is implicitly displaced. Neither has happened. The framework advocates for the former. As legal analysts at JD Supra noted in reviewing the framework, the recommendation is legislative, not operative. Congress would need to pass a bill. That bill does not yet exist.
Senator Marsha Blackburn (R-TN) announced the TRUMP AMERICA AI Act discussion draft in the same week, a document described as intended to codify the president’s AI preemption executive order. This draft is a discussion document, not an introduced bill. It is the first concrete legislative vehicle for the administration’s preemption agenda, which makes it worth tracking, but not worth building compliance decisions around.
Washington State’s laws, by contrast, are operative. They were signed. They have effective dates. They will be enforced.
The Compliance Problem: Simultaneous Obligations Under Conflicting Signals
Here’s the situation AI developers actually face: Washington State has enacted real obligations. The federal government has signaled it wants those obligations preempted. The preemption might happen, eventually. It might not. It certainly hasn’t happened yet.
Compliance teams navigating this gap should think about it in three layers.
Layer one: current obligations. Washington State’s chatbot safety and AI content disclosure requirements apply now to covered entities. The detection tool requirement is reportedly set to take effect in 2027, according to initial reporting, though the specific date and the “large AI developers” threshold definition require confirmation against the official enrolled bill text (available at leg.wa.gov). Until that text is reviewed, compliance timelines should be treated as approximate. The obligation itself is not in question.
Layer two: preemption risk assessment. Organizations should map their current state AI compliance portfolio, not just Washington State, but any state where AI-specific obligations exist or are pending. This mapping becomes the input to a preemption analysis if and when federal legislation advances. The question to answer: which of my current state law obligations would be affected by preemption, under what carve-outs, and with what transition timelines? You cannot answer that question without first knowing what you’re complying with today.
Layer three: the uncertainty premium. Legal analysts reviewing the framework report that proposed carve-outs would preserve state authority over areas including child protection, consumer protection, and fraud. The exact scope of those carve-outs would be defined by legislation that doesn’t exist yet. This uncertainty is not resolvable from the framework text alone. Build for current requirements. Flag preemption exposure as a monitoring item, not an action item.
Who Is Affected, and How
The affected population is wider than Washington State alone.
AI developers with Washington State exposure: Any developer whose AI products are deployed in Washington State and who meets whatever size threshold HB 2225 defines needs to be tracking the bill text, the effective dates, and the specific technical requirements for detection tools. The chatbot safety provisions are relevant for any developer offering conversational AI that could interact with minors.
AI developers with multi-state compliance programs: The Washington signing is a signal, not an outlier. If you’re managing compliance across multiple state AI laws, the framework’s preemption recommendation should prompt you to document the cost and complexity of your current multi-state obligations. That documentation becomes valuable both for internal planning and, potentially, as input to policy engagement if Congress opens a comment process.
General counsel and government affairs teams: The Blackburn discussion draft is the legislative vehicle to watch. Its progress from draft to introduced bill to committee action is the signal that federal preemption has moved from policy aspiration to legislative reality. Track it. Engage with the process when it opens.
State-level compliance programs that were built around state law flexibility: If your organization built a compliance program intentionally tailored to state-level requirements, treating state specificity as a feature, federal preemption would require structural rethinking. That’s not an imminent problem, but it’s worth identifying now rather than scrambling to answer it when a bill clears committee.
What to Watch
Three specific triggers warrant escalating attention.
First: formal introduction of the Blackburn TRUMP AMERICA AI Act. A discussion draft advancing to a numbered bill with committee assignment is the first concrete signal that the legislative timeline is real.
Second: the official text of HB 2225 and Washington State’s companion bill. The enrolled bill text, available through the Washington State Legislature at leg.wa.gov, will confirm effective dates, threshold definitions, and technical compliance requirements. The next Wire cycle should retrieve this directly.
Third: whether other states follow Washington’s pattern. If states continue enacting AI laws while Congress debates preemption, the gap between federal intent and state action widens. Each new state law adds to the complexity that preemption is supposed to resolve, and adds to the argument that Congress needs to move faster.
The Signal Beneath the Noise
The same-day collision between the White House framework and the Washington State signing isn’t a coincidence worth noting and moving past. It’s the defining condition of U.S. AI governance right now: a federal government with a clear policy direction and no legislative vehicle to implement it, and a growing number of state governments acting in the absence of federal action.
For AI developers, the practical implication is this: you are operating in the gap. Federal preemption, if it comes, will change the rules. Until it does, the rules are whatever the states have enacted. Build for that reality. Monitor the other.