A National Policy Framework for Artificial Intelligence: Legislative Recommendations is a set of legislative recommendations released by the White House on March 20, 2026, outlining the Trump administration's proposed approach to federal regulation of artificial intelligence. The four-page document calls on Congress to enact legislation covering seven areas: child safety, community protections, intellectual property, free speech, innovation, workforce development, and federal preemption of state AI laws. The document is non-binding and does not carry the force of law; implementing its proposals would require congressional action. The framework was developed by OSTP Director Michael Kratsios and White House Special Advisor for AI and Crypto David Sacks, who indicated the administration intended to work with Congress to turn the recommendations into law within the year.
The framework followed a series of executive actions on AI policy during the second Trump administration. In January 2025, Trump signed Executive Order 14179, revoking the Biden administration's October 2023 executive order on AI (Executive Order 14110) and directing federal agencies to remove what the administration characterized as regulatory barriers to AI development. In July 2025, the administration published a policy blueprint titled "Winning the AI Race: America's AI Action Plan."
On December 11, 2025, Trump signed an executive order titled "Ensuring a National Policy Framework for Artificial Intelligence," which took more aggressive steps toward displacing state AI regulation. The order directed the Department of Justice to establish an AI Litigation Task Force to challenge state AI laws in court, instructed the Department of Commerce to identify state laws deemed "onerous" by March 2026, and threatened to withhold BEAD program funding from states whose AI laws the administration viewed as conflicting with federal policy. That order also directed Sacks and Kratsios to prepare the legislative recommendations that were ultimately released on March 20, 2026.
Legal commentators noted that the December 2025 executive order, on its own, lacked preemptive force because it was neither a statute enacted by Congress nor a regulation issued pursuant to congressional authorization. Law firms including Latham & Watkins, Ropes & Gray, and Paul Hastings advised clients that state AI laws remained enforceable pending either congressional action or successful litigation.
Prior to the March 2026 framework, the administration and its congressional allies had attempted several times to preempt state AI regulation through legislation, without success. In mid-2025, a proposed 10-year moratorium on new state AI laws was included in a reconciliation bill, but the Senate voted 99âÂÂ1 to strip the provision following bipartisan opposition. A similar effort to include preemption language in the National Defense Authorization Act for fiscal year 2026 also failed.
By March 2026, four states had enacted laws setting private-sector requirements for AI: Colorado, California, Utah, and Texas. Colorado's AI Act, which went into effect in February 2026 and required developers and deployers of high-risk AI systems to exercise "reasonable care" to protect consumers from algorithmic discrimination, was cited specifically in the December 2025 executive order as an example of problematic state regulation. More than 1,000 AI-related bills had been introduced across state legislatures by this time.
The administration's preemption efforts had drawn opposition from state officials across party lines. More than 280 state lawmakers from both parties signed a letter urging Congress to reject any legislation preempting state AI regulation, and a bipartisan coalition of 36 state attorneys general issued similar objections. In response to the December 2025 executive order, Florida Governor Ron DeSantis argued on social media that executive orders cannot preempt state legislation and that only Congress has that authority. The Center for American Progress described the preemption effort as proposing to shield AI companies from state regulation without providing any meaningful substitute.
The administration's approach also reflected internal tensions. CNN reported that within the White House, Sacks and Mike Davis, a senior aide to Vice President JD Vance, had clashed over the pace and scope of AI deregulation, with Davis accusing Sacks of attempting to impose the technology on the country without adequate safeguards. Former Trump adviser Steve Bannon publicly opposed the administration's pro-industry posture, stating after the December 2025 executive order that opponents would "fight like hell." A TIME analysis noted a growing populist backlash against data centers and AI-driven economic disruption, particularly in swing states.
Earlier stages of the administration's AI policy had drawn criticism from policy analysts. The Brookings Institution, analyzing the July 2025 AI Action Plan, criticized the administration's approach for giving insufficient attention to accountability, ethics, and transparency, and for lacking a domestic competition policy to prevent concentration among a small number of dominant AI firms. The December 2025 executive order had directed the FTC to issue a policy statement characterizing state-mandated bias mitigation as potentially deceptive under federal law, on the theory that requiring AI models to adjust outputs to reduce discriminatory impact would make those outputs less "truthful"; Paul Hastings noted that courts might reject the premise that correcting for bias constitutes deception.
The framework is organized into seven sections, each containing recommendations to Congress.
The first section calls on Congress to give parents tools to manage children's privacy, screen time, and content exposure on AI platforms. It recommends "commercially reasonable" age-assurance requirements, such as parental attestation, for AI services likely to be accessed by minors. It also recommends that platforms implement features to reduce risks of sexual exploitation and self-harm to minors. The framework references the TAKE IT DOWN Act, signed into law on May 19, 2025, which criminalized the non-consensual distribution of intimate images including AI-generated deepfakes. The section also warns against setting ambiguous content standards that could generate excessive litigation and specifies that federal legislation should not prevent states from enforcing generally applicable child protection laws, including prohibitions on AI-generated child sexual abuse material.
The second section addresses energy costs, infrastructure permitting, fraud, national security, and small business support. It calls on Congress to codify the Ratepayer Protection Pledge, a voluntary commitment signed on March 4, 2026, by Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI, in which those companies agreed to cover the full costs of electricity generation and infrastructure upgrades for their data centers, rather than passing costs to residential utility customers. The pledge was first announced during Trump's State of the Union address on February 24, 2026. The framework also recommends streamlined federal permitting for on-site power generation at AI facilities, strengthened law enforcement tools to combat AI-enabled impersonation scams targeting seniors, and increased technical capacity within national security agencies to evaluate frontier AI model capabilities.
The IP section adopts a position that training AI models on copyrighted material does not violate copyright law, but acknowledges that contrary arguments exist and recommends that Congress let the courts resolve the question rather than intervening legislatively. This position was consistent with remarks Trump had made at a July 2025 AI summit, where he characterized requiring AI companies to pay for all training data as "not doable." The Associated Press reported that the framework effectively deferred to the courts on the question of fair use, noting that dozens of lawsuits from writers, publishers, visual artists, and music labels were pending on the issue.
The section also suggests Congress consider enabling licensing frameworks or collective rights systems through which rights holders could negotiate compensation from AI providers without incurring antitrust liability, while specifying that any such legislation should not determine when or whether licensing is required. It further recommends a federal framework protecting individuals from unauthorized commercial use of AI-generated replicas of their voice or likeness, with exceptions for parody, satire, and news reporting.
The fourth section calls on Congress to prevent the federal government from pressuring AI providers to alter or suppress content based on partisan or ideological agendas. It recommends providing a legal mechanism for individuals to seek redress if federal agencies attempt to censor expression on AI platforms or dictate the information those platforms provide. Bloomberg News noted that this provision reflected longstanding conservative allegations that technology companies exhibited bias against their political views, which the companies have denied.
The fifth section calls for the establishment of regulatory sandboxes for AI applications, making federal datasets available in formats suitable for AI training, and opposing the creation of any new federal rulemaking body dedicated to AI. It instead recommends that existing sector-specific regulators and industry-led standards govern AI deployment.
The sixth section recommends incorporating AI training into existing education, workforce development, and apprenticeship programs. It calls for expanded federal research into task-level workforce changes driven by AI and increased support for land-grant institutions to develop AI-related technical assistance, demonstration projects, and youth programs.
The final section, described by Axios as the framework's most consequential recommendation, calls on Congress to preempt state AI laws that the administration considers unduly burdensome, replacing them with a single national standard. The framework carves out several categories of state law that it says should not be preempted: traditional police power to enforce laws of general applicability (including child protection, fraud prevention, and consumer protection), state zoning laws governing AI infrastructure placement, and requirements governing a state's own use of AI through procurement or public services such as law enforcement and education.
At the same time, the framework argues that states should not be permitted to regulate AI development, characterizing it as "an inherently interstate phenomenon with key foreign policy and national security implications." It also recommends that states not be allowed to penalize AI developers for unlawful conduct by third parties using their models, and that states not impose additional burdens on AI-assisted activities that would be lawful if performed without AI.
The framework calls on Congress to preempt state AI laws it considers burdensome, but does not itself propose specific federal requirements to replace the state-level protections that would be displaced. It also contains no provisions addressing algorithmic discrimination or bias in AI systems, topics that had been central to state-level AI legislation in Colorado, California, and elsewhere.
In a press release accompanying the framework, Kratsios stated that the proposal would "unleash American ingenuity to win the global AI race" while addressing concerns about child safety, energy costs, intellectual property, and workforce readiness. In an interview with Fox News, Kratsios and Sacks stated they had already consulted with House and Senate leadership and expressed confidence the framework could attract bipartisan support.
The Center for Data Innovation praised the framework, with Director Daniel Castro stating that it "recognizes that the United States will not lead in AI by regulating from a place of fear" and that it addressed real harms without impeding innovation. Civil liberties and consumer rights organizations had generally been pushing for stricter AI regulation, and the Associated Press reported that these groups lobbied for more oversight of the technology than the framework proposed.
Bloomberg News described Congress as "skeptical" of the plan. Axios noted that disagreements over federal preemption, copyright, and child safety had been the same issues that stalled congressional action on AI for years, and that the push to override state laws would continue to face resistance from state officials in both parties. CNBC reported that the effort faced additional obstacles in the form of narrow Republican congressional majorities and the administration's prioritization of a voter-ID bill ahead of the 2026 midterm elections. A Roll Call analysis published weeks before the framework's release noted that while there was broad agreement on the desirability of a national standard, experts were skeptical about the prospects for passage in 2026, citing thin Republican margins and a busy election-year calendar.
Two days before the framework's release, Senator Marsha Blackburn (R-TN) had released a discussion draft of legislation titled the "TRUMP AMERICA AI Act" intended to codify key elements of the December 2025 executive order. Blackburn's draft diverged from the administration's position on copyright: it would have established that unauthorized use of copyrighted works in AI training does not constitute fair use, contradicting the administration's stated preference for letting the courts decide.