Last reviewed
Jun 9, 2026
Sources
10 citations
Review status
Source-backed
Revision
v2 ยท 2,737 words
Add missing citations, update stale details, or suggest a clearer explanation.
The National Policy Framework for Artificial Intelligence is a White House document, subtitled "Legislative Recommendations," that the administration of President Donald Trump released on March 20, 2026. It sets out a list of principles the administration wants Congress to use as the basis for federal artificial intelligence legislation, organized into seven thematic areas. The most contested of these is a call for federal preemption of state AI laws, which the document frames as necessary to avoid "fifty discordant" standards while preserving traditional state powers in areas such as child protection, fraud, consumer protection, and zoning. The framework is a short statement of policy positions, roughly four pages long, and does not contain draft statutory text or create any binding legal obligation on its own.[1][2][3]
The document was prepared in response to a directive in Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," which President Trump signed on December 11, 2025. That order instructed the Special Advisor for AI and Crypto, David Sacks, together with the Assistant to the President for Science and Technology and Director of the Office of Science and Technology Policy, Michael Kratsios, to prepare legislative recommendations for a uniform federal AI framework that would preempt conflicting state laws. The March 2026 framework is the product of that instruction. It builds on America's AI Action Plan, the administration's July 2025 strategy document, and shares its emphasis on innovation, energy infrastructure, and removing regulatory barriers.[1][4][5]
This article describes the framework document itself: its origin, its seven recommended policy areas, the preemption proposal and its carve-outs, how it differs from the executive order that produced it, and the early reaction in Congress and among analysts. The executive order, the litigation task force it created, and the federalism litigation strategy are treated in the separate article on Executive Order 14365.
The framework is a set of recommendations addressed to Congress rather than a regulation or an executive action with its own legal force. Several law firm analyses stressed this point: the document provides "recommended policy approaches for Congress to consider," is "not binding," and "does not, on its own, create new legal obligations."[2][3] Because it contains no draft bill, its practical effect depends on whether members of Congress translate its principles into legislation and whether such legislation passes.
The White House announced the framework through a release titled "President Donald J. Trump Unveils National AI Legislative Framework," which described the goal as "winning the AI race" while ensuring that "all Americans benefit from this technological revolution." The administration urged Congress to act in 2026.[6][7] The document closes the loop on a sequence that began with the AI Action Plan in July 2025 and the December 2025 executive order: the order set federal policy and directed agencies to push back against state laws, and the framework supplies the legislative blueprint the order asked for.[4][1]
The administration's approach to AI was laid out in America's AI Action Plan, released on July 23, 2025, which treated rapid AI development as a national priority and called for removing regulatory barriers to it. The deregulatory orientation of that plan carried through to the December 2025 executive order and to the March 2026 framework, all of which favor a light-touch federal posture and centralized rather than state-by-state rules.[5][4] For broader context on presidential AI directives, see Executive Order on AI.
Executive Order 14365, signed December 11, 2025, declared a national policy of sustaining United States "global AI dominance through a minimally burdensome national policy framework," directed federal agencies to challenge state AI laws the administration considered onerous, and asked David Sacks and Michael Kratsios to draft legislative recommendations for a uniform federal standard with preemption. The March 2026 framework is that deliverable.[1][4]
A core premise of the framework is that a growing number of differing state AI laws creates a fragmented compliance landscape. The administration and industry groups argue that dozens of separate state regimes covering transparency, automated decision systems, deepfakes, and developer obligations raise costs and slow national deployment, with the burden falling hardest on smaller developers. Examples frequently cited in the debate include California Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act; the Colorado Artificial Intelligence Act; and the Texas Responsible AI Governance Act. Opponents read the same proliferation of state laws as evidence that states are responding to real problems, and warn that removing those laws without enacting federal protections would leave a gap.[2][8]
Two earlier efforts to preempt state AI law through legislation had failed before the framework appeared. A ten-year moratorium on state AI enforcement passed the House as part of a 2025 budget reconciliation bill but was stripped by the Senate in a 99 to 1 vote on July 1, 2025. A later attempt to attach preemption language to the annual defense authorization bill also failed in late 2025. The executive order, and then the framework, followed those defeats.[1][8]
The document organizes its recommendations into seven numbered sections. Each opens with a one-sentence statement of principle followed by specific suggestions for Congress. The table below lists the sections as titled in the document, with a short summary of each.[1][9]
| # | Section title | Core recommendations |
|---|---|---|
| I | Protecting Children and Empowering Parents | Build on the Take It Down Act; give parents tools over privacy, screen time, content, and account controls; establish "commercially reasonable, privacy protective, age-assurance requirements (such as parental attestation)" for AI platforms likely accessed by minors; reduce risks of sexual exploitation and self-harm; avoid ambiguous content standards or open-ended liability that invites litigation |
| II | Safeguarding and Strengthening American Communities | Protect residential ratepayers from higher electricity costs caused by AI data centers under a "Ratepayer Protection Pledge"; streamline federal permitting for AI infrastructure and behind-the-meter power; combat AI-enabled impersonation scams targeting seniors; ensure national security agencies can assess frontier model capabilities; provide grants, tax incentives, and technical assistance to small businesses |
| III | Respecting Intellectual Property Rights and Supporting Creators | State the administration's view that training on copyrighted material does not violate copyright law while leaving the fair use question to the courts; consider licensing or collective rights frameworks without antitrust liability; consider a federal right against unauthorized AI-generated "digital replicas" of voice or likeness, with First Amendment exceptions for parody, satire, and news |
| IV | Preventing Censorship and Protecting Free Speech | Bar the federal government from coercing technology and AI providers to ban, compel, or alter content for partisan or ideological reasons; give Americans a means of redress against federal efforts to censor expression on AI platforms |
| V | Enabling Innovation and Ensuring American AI Dominance | Establish regulatory sandboxes for AI applications; make federal datasets available in AI-ready formats; decline to create any new federal AI rulemaking body, relying instead on existing sector regulators and industry-led standards |
| VI | Educating Americans and Developing an AI-Ready Workforce | Use non-regulatory means to add AI training to existing education, workforce, and apprenticeship programs; expand federal study of AI-driven workforce realignment; strengthen land-grant institutions and AI youth development programs |
| VII | Establishing a Federal Policy Framework, Preempting Cumbersome State AI Laws | Preempt state AI laws that impose undue burdens to create "a minimally burdensome national standard," while respecting federalism and preserving specified state powers |
The recommendation against creating a new federal AI regulator is notable for departing from approaches that build a dedicated agency. The document states that Congress "should not create any new federal rulemaking body to regulate AI" and should instead work through existing sector-specific regulators and industry-led standards, an approach consistent with the deregulatory frame of the AI Action Plan and with the NIST AI Risk Management Framework model of voluntary guidance.[1][2]
The seventh section is the framework's most debated element. It asks Congress to "preempt state AI laws that impose undue burdens to ensure a minimally burdensome national standard consistent with these recommendations, not fifty discordant ones." The document argues that states "should not be permitted to regulate AI development, because it is an inherently interstate phenomenon with key foreign policy and national security implications," that states should not "unduly burden Americans' use of AI for activity that would be lawful if performed without AI," and that states should not "penalize AI developers for a third party's unlawful conduct involving their models."[1][9]
At the same time, the framework explicitly limits what preemption should reach, citing "key principles of federalism." It states that a national standard should not preempt the traditional police powers states retain to enforce laws of general applicability, including laws to protect children, prevent fraud, and protect consumers; state zoning laws governing the placement of AI infrastructure; or requirements governing a state's own use of AI, whether in procurement or in services such as law enforcement and public education.[1][3] In the child-safety section, the document separately stresses that any federal legislation should not preempt states from enforcing generally applicable laws protecting children, "such as prohibitions on child sexual abuse material, even where such material is generated by AI."[1]
The table below summarizes the preemption recommendation as the document frames it.
| What the framework would preempt | What it would preserve for states |
|---|---|
| State laws regulating AI development as an interstate activity | Police powers and laws of general applicability (child protection, fraud, consumer protection) |
| State laws unduly burdening lawful uses of AI | State zoning and placement of AI infrastructure |
| State laws penalizing developers for third parties' unlawful conduct | A state's own use of AI in procurement and public services |
Analysts noted that the boundary between preempted and preserved categories is left undefined. A review by the Center for Security and Emerging Technology observed that the document does not make clear whether its preemption language would cover high-risk uses such as employment and healthcare, leaving significant ambiguity for Congress and the courts to work out.[9]
The framework is a product of, and is distinct from, Executive Order 14365. The executive order is a binding directive that, among other things, created a Department of Justice AI Litigation Task Force to challenge state AI laws in court, directed the Department of Commerce to evaluate conflicting state laws, conditioned certain broadband funding on a state's AI regulatory climate, and tasked Sacks and Kratsios with preparing legislative recommendations. The March 2026 framework is the legislative-recommendation deliverable from that last instruction. It carries no legal force of its own and works only by persuading Congress.[1][4]
The two documents share a title phrase, "National Policy Framework for Artificial Intelligence," which has caused some confusion. The executive order is named "Ensuring a National Policy Framework for Artificial Intelligence," while the March 2026 document is titled "A National Policy Framework for Artificial Intelligence: Legislative Recommendations." The first is an executive action signed in December 2025; the second is a non-binding set of recommendations released in March 2026. They are complementary parts of the same overall strategy: the order uses executive tools available without new legislation, and the framework sets out what the administration wants that legislation to say.[1][3]
The framework also fits within the wider arc of the administration's AI policy that began with America's AI Action Plan. The plan, the order, and the framework together describe a strategy of pairing a deregulatory federal posture with pressure against state regulation, and then asking Congress to lock in a single national standard with preemption.[5][4]
Reaction in Congress tracked the framework's release closely. Two days before the document came out, on March 18, 2026, Senator Marsha Blackburn released an updated discussion draft of a bill styled the "TRUMP AMERICA AI Act," which several analyses described as closely tracking the framework's themes while adding provisions such as a duty of care for AI chatbot developers and child-online-safety requirements. Whether Senate Commerce Committee Chairman Ted Cruz would support the preemption strategy was reported as uncertain.[2][3]
On the day of the release, March 20, 2026, a group of House Democrats led by Representative Don Beyer, joined by Representatives Doris Matsui, Ted Lieu, Sara Jacobs, and April McClain Delaney, introduced the GUARDRAILS Act, which would repeal Executive Order 14365 and block efforts to impose a moratorium on state-level AI regulation. The bill was framed as ensuring that states can continue to adopt laws to protect the public.[2][10]
Policy analysts gave the framework a mixed assessment. The Center for Security and Emerging Technology identified several gaps: it found that frontier AI risks received minimal attention relative to their prominence in the broader debate, that the document offered little specificity on how regulatory sandboxes would work, and that AI literacy initiatives were largely absent from the workforce section. The same analysis suggested the prominent child-protection provisions might reflect either genuine bipartisan common ground or a strategic effort to soften political resistance to preemption.[9] Several law firms framed the document as a starting point, predicting that near-term legislative progress was more likely in discrete areas such as child safety, transparency, and fraud prevention than in a single comprehensive bill, given opposition to broad preemption on both sides of the aisle.[3][2]
The framework matters mainly as the administration's formal statement of what it wants federal AI legislation to contain. After two failed attempts to preempt state AI law through Congress in 2025, and an executive order that can pressure but not by itself displace state statutes, the framework is the administration's attempt to set the terms of the legislative debate that durable preemption would require. By packaging preemption alongside child safety, free speech, intellectual property, and workforce measures, it offers Congress a menu from which individual provisions might advance even if the central preemption proposal stalls.[2][9]
The document also clarifies the administration's preference on institutional design. Its explicit recommendation against any new federal AI regulator, in favor of existing sector regulators and industry-led standards, distinguishes the United States approach from the centralized model of the EU AI Act and aligns it with the voluntary, framework-based tradition reflected in the NIST AI Risk Management Framework. That contrast feeds the wider debate in AI regulation and AI governance over whether AI is best governed through a single national rule, a dedicated agency, or sectoral oversight.[1][9]
As of mid-2026, the framework remained a set of recommendations rather than enacted law. No comprehensive federal AI statute reflecting it had passed, and the central preemption proposal faced opposition from Democrats and from some Republicans who defend state authority. Competing bills were in circulation, including Senator Blackburn's discussion draft on one side and the GUARDRAILS Act on the other, and analysts expected stakeholder engagement to shape any statutory language that emerged.[3][10] In the meantime, the executive order's mechanisms continued to operate in parallel, and state AI laws such as California's SB 53 remained in effect, so companies faced a continuing mix of state and federal obligations while the legislative debate played out.[2][9] Whether the framework's principles become law, and in what form, will help determine whether AI in the United States is governed primarily from Washington or from state capitals.