The week of March 20, 2026, produced two significant regulatory moves in opposite directions.
On March 20, the Trump Administration released a National Policy Framework for AI, calling on Congress to preempt state laws the administration deems “unduly burdensome” to innovation. Six days later, on March 26, the European Parliament voted to extend three major EU AI Act compliance deadlines and add a ban on AI nudifier systems to the regulation’s scope. Two of the world’s most significant AI regulatory frameworks moved in the same week. They moved in opposite directions.
For organizations with exposure in both markets, that divergence isn’t theoretical. It’s a planning problem.
Section 2: What the US Framework Actually Does, and Doesn’t Do
The US framework is a legislative recommendation, not a law. That distinction matters more than the framework’s substance.
The White House framework, released pursuant to Executive Order 14365 from December 2025, identifies seven priority areas for federal AI legislation: child protection, community and infrastructure impacts, intellectual property, free speech, innovation, workforce development, and federal preemption of state AI laws. The preemption mechanism is the headline provision. Its logic: a patchwork of state AI laws undermines innovation and creates compliance complexity for companies operating across state lines.
The administration has also taken a position on AI copyright. The framework states the administration’s view that “training of AI models on copyrighted material does not violate copyright laws,” while recommending Congress avoid legislation that would interfere with ongoing judicial proceedings on the question. That’s a position, not a ruling. Courts are still deciding.
Two bills have emerged from this framework. Senator Marsha Blackburn (R-TN) released a discussion draft of the TRUMP AMERICA AI Act on March 18, described by analysts as among the most comprehensive federal AI legislative proposals introduced to date. House Democrats responded with the GUARDRAILS Act, introduced in direct response to the preemption provisions, according to reporting from Let’s Data Science. The GUARDRAILS Act’s specific provisions weren’t confirmed at publication time.
What this means practically: no state AI law has been preempted. California, Colorado, and a growing cohort of states with active AI legislation remain in effect. Washington State, for instance, has enacted AI chatbot safety and deepfake disclosure laws that are currently operative. Federal preemption requires both bills to move through Congress, and the GUARDRAILS Act signals that movement faces organized opposition. That process hasn’t started in any meaningful sense. The framework and the Blackburn bill are starting points, not outcomes.
The timeline uncertainty is significant. Organizations should not assume that the US deregulatory posture will translate into near-term legal relief from state AI compliance obligations. The more conservative planning assumption is that state laws govern until they don’t, and the legislative process needed to change that is contested and uncertain.
Section 3: What the EU Deadline Shift Actually Changes
The EU picture is more concrete, because the European Parliament is working within an existing legal framework rather than proposing to build one.
The Parliament adopted its position on a simplification proposal for the EU AI Act on March 26. Three implementation deadlines are proposed to shift:
| Compliance Area | Previous Timeline | Proposed New Deadline | Status |
|---|---|---|---|
| AI-generated content watermarking | [Prior deadline] | November 2, 2026 | Proposed, pending Council approval |
| High-risk AI systems (Annex III) | [Prior deadline] | December 2, 2027 | Proposed, pending Council approval |
| AI systems under sectoral safety law | [Prior deadline] | August 2, 2028 | Proposed, pending Council approval |
The Parliament stated the delays are intended to ensure that guidance and technical standards are in place before companies must comply, an acknowledgment that the original timeline outpaced the infrastructure needed to meet it.
The nudifier ban is a different kind of action. Parliament’s vote didn’t just extend deadlines; it also proposed a new prohibition on AI systems that create or manipulate sexually explicit images resembling identifiable real persons without consent. This is a scope expansion, not a delay. The regulation is growing in coverage at the same time it’s growing in timeline.
One critical procedural point: none of these changes are final. Parliament’s position now goes to the Council of the EU. Until the Council approves, these proposed dates and the proposed nudifier ban are not binding obligations. Compliance teams should plan against them, but the formal effective dates require Council action.
Member state implementation readiness varies. Several EU member states have yet to designate national enforcement contacts under the Act’s requirements. That gap matters because enforcement under the EU AI Act ultimately depends on national competent authorities, and the infrastructure for enforcement is still being built across the union. The Parliament’s decision to extend deadlines reflects awareness of this reality.
Section 4: The Compliance Divergence in Practice
An organization operating in both the US and EU markets now faces a split planning environment.
On the US side: the federal direction is deregulatory, but the legal reality is unchanged. State laws are in effect. Federal preemption is a legislative goal with named bills and named opposition. The administration’s position on copyright is stated, not settled. For compliance purposes, the safest assumption is that current state law obligations continue until Congress acts, and Congressional action is neither certain nor imminent.
On the EU side: the legal framework exists and is being refined. The proposed deadline extensions provide runway, but the watermarking deadline is only seven months away. November 2, 2026 is the nearest obligation on the revised calendar. Any provider generating AI audio, image, video, or text for EU audiences should be scoping watermarking implementation now. The extended high-risk and sectoral deadlines (December 2027 and August 2028) provide more time, but “more time” isn’t “no obligation.” The regulation exists. Compliance planning should be underway.
The practical divergence looks like this: US-facing compliance teams may be tempted to interpret the administration’s deregulatory posture as a signal to deprioritize state law compliance. That interpretation is premature. EU-facing compliance teams may read the deadline extensions as breathing room to slow planning. The watermarking deadline argues otherwise.
The organizations best positioned for what comes next are those that track both tracks independently. The US legislative outcome is genuinely uncertain. The EU timeline is extended but moving. Neither situation rewards an assumption that the pressure has eased.
Section 5: What to Watch Next
Four milestones shape how this story develops.
The Council of the EU’s response to Parliament’s position is the immediate one. Until the Council acts, the proposed EU deadline dates remain proposals. The Council could approve, modify, or reject Parliament’s simplification proposal. Watch for a Council decision on the proposed watermarking date especially, November 2026 gives organizations limited time to respond once the date is confirmed.
On the US side, committee assignments for the Blackburn discussion draft will signal whether it advances beyond the discussion stage. The GUARDRAILS Act’s bill text and sponsors, when confirmed, will clarify the scope of Democratic opposition. Both pieces of information are pending.
Congressional action on federal preemption, any action, is the milestone that would actually change the US state-law compliance environment. Without it, state laws remain the operative legal framework for AI regulation in the US. Watch the legislative calendar, not the executive rhetoric.
A third US data point may materialize soon. A federal court case involving potential government restrictions on Anthropic is under active research and will be covered in a forthcoming brief when sources are confirmed. If verified, it would add a judiciary dimension to the US regulatory picture, complementing the legislative and executive dynamics covered here.
The synthesis. Both moves this week reveal something about where their respective systems are. The US framework’s ambition is large; its legal force is currently zero. The EU framework’s ambition is also large, and it has the force of existing law, but its implementation timeline has moved twice now, and the infrastructure for enforcement isn’t fully in place. Neither regulatory environment is stable. Both require active monitoring. Organizations managing compliance in both markets are doing so in a period when the rules are being negotiated on both sides of the Atlantic, simultaneously, in different directions.
That’s the environment. Plan accordingly.