The White House framework for a national AI law: what it proposes and what it changes today

In short: the National Policy Framework for Artificial Intelligence, released by the White House on March 20, 2026, is four pages of non-binding legislative recommendations to Congress. It proposes a federal standard that would preempt “unduly burdensome” state AI laws, limits on developer liability, child-safety measures and a view on copyright. It changes no one’s obligations today; state laws remain in force until Congress legislates.^1,2

The framework does not create legal obligations or direct agencies to act.¹ It follows Executive Order 14365 of December 2025, which asked advisers to prepare exactly this kind of proposal for a federal framework that would preempt conflicting state laws.³ This briefing is educational, not legal advice.

What it recommends

The framework is organized around seven areas.¹ The table sets out what each would mean for an organization that deploys AI, if Congress adopted it.

The framework relies on existing agencies rather than a new AI regulator.¹ McGuireWoods’ reading adds what is missing: it is largely silent on enforcement, proposes no specific penalties or compliance mechanisms, does not create a comprehensive privacy regime, and does not address AI-driven discrimination. Whether Congress can pass comprehensive legislation at all is an open question.²

What changes today

Nothing, legally. Recommendations do not preempt anything; only Congress can do that by statute, and in July 2025 the Senate voted 99 to 1 against a ten-year moratorium on state AI laws.⁴ State laws in force, such as Texas’s AI governance act and Illinois’s rules on AI in employment decisions, remain enforceable, and Colorado’s AI Act was, at the time, scheduled for June 30, 2026 (it has since been repealed and replaced by SB 26-189, which applies from January 1, 2027; see the update note above). Our briefing on the January 1, 2026 laws covers what took effect this year.

How we got here

• July 1, 2025: the Senate removed a proposed ten-year moratorium on state AI laws from the budget bill, 99 to 1.⁴
• December 11, 2025: Executive Order 14365 directed a litigation task force to challenge state AI laws the administration considers burdensome, conditioned some federal funding, and asked for a legislative proposal.³
• March 20, 2026: the White House released the proposal as this framework, going further than the order by asking Congress for statutory preemption.^1,2

Each step has moved the question from the executive branch toward Congress. The order can pressure states; only a statute can displace their laws.

How to plan

Two things follow for compliance teams. First, keep building to the state rules that apply now; a federal statute, if one passes, will take time, and its preemption scope is unknown. Second, note the liability recommendation. If developers’ liability for downstream use is limited, more of the responsibility for how a system is used sits with the organization that deploys it. That makes three deployer habits more valuable under either outcome:

1. An inventory of AI systems with the people and decisions each one affects.
2. Impact assessments and human review for consequential decisions such as hiring, lending and access to services.
3. Contracts that keep vendor commitments on testing, documentation and indemnity, because the law may not.

What to watch next

Three signals will matter more than the framework itself: whether a bill carrying its preemption language is introduced and gets a committee hearing; the Commerce Department’s evaluation of state AI laws and any suits filed by the Justice Department’s AI Litigation Task Force under the executive order; and whether states keep legislating regardless. Put each on the regulatory watch list your team reviews every month, with a named owner, rather than reacting to headlines.³

The US federal posture and the state laws are taught side by side in Module 6 of the AIMP program. Strong deployer governance is the posture that holds up whichever way Congress goes.