Trump’s AI Executive Orders Explained: EO 14179 and EO 14365

Two Trump administration executive orders now define federal AI policy: Executive Order 14179, “Removing Barriers to American Leadership in Artificial Intelligence” (signed January 23, 2025), and Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence” (signed December 11, 2025). Neither is a law passed by Congress, and neither directly regulates what an AI company must do — but together they set the federal government’s posture toward the state AI laws and the EU AI Act that do impose binding requirements. This guide explains what each order actually says, primary-source verified against the Federal Register, and how the two relate to the rest of the AI regulatory landscape.

What a federal executive order actually is (and isn’t)

An executive order is a directive the President issues to the executive branch under existing constitutional or statutory authority. It is not legislation: Congress does not vote on it, and unlike a statute it can be revoked by a later President with a single signature, exactly as EO 14179 revoked its predecessor. An executive order also cannot, on its own, create new legal obligations for private companies the way a statute or a duly-issued regulation can — it can only direct federal agencies, and any binding rule that results has to come from the ordinary channels available to those agencies (rulemaking, existing statutory authority such as the Defense Production Act, litigation, or funding conditions attached to programs Congress already created). That distinction matters for both orders covered here: as written, neither EO 14179 nor EO 14365 imposes a new legal duty directly on an AI developer. Both work by steering what federal agencies do next.

EO 14179: Removing Barriers to American Leadership in Artificial Intelligence

Signed January 23, 2025 and published in the Federal Register on January 31, 2025, EO 14179 does two things. First, Section 3 revokes Executive Order 14110, “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence” — the Biden administration’s October 2023 AI order — in full. Second, it directs a new policy of sustaining and enhancing U.S. AI dominance, free of what it characterizes as ideological bias or engineered social agendas, and sets deadlines for federal agencies to act on that policy:

• Immediately: the Assistant to the President for Science and Technology (APST), the Special Advisor for AI and Crypto, and the Assistant to the President for National Security Affairs (APNSA) were directed to review all policies, directives, regulations, orders and other actions taken under the revoked EO 14110 and identify anything inconsistent with the new policy.
• Within 60 days: the OMB Director was directed to revise OMB Memoranda M-24-10 and M-24-18 — the federal government’s own internal AI governance and procurement guidance — to align with the new policy.
• Within 180 days: the APST, Special Advisor for AI and Crypto, and APNSA were directed to develop and submit an AI action plan to the President. That deadline produced the administration’s “America’s AI Action Plan,” released in July 2025.

What EO 14179 does not do is create any new requirement that an AI company must meet. Its practical effect on companies is the opposite: it eliminates one. EO 14110 had used the Defense Production Act to require developers of the largest “dual-use foundation models” — defined by a compute threshold of greater than 10²⁶ integer or floating-point operations — to report to the federal government on their training activities, the security of their model weights, and the results of their AI red-team safety testing. EO 14110 also directed NIST to publish red-teaming guidelines for exactly this kind of testing. Revoking EO 14110 ended that federal reporting requirement; nothing in EO 14179 replaces it with a requirement of its own.

EO 14365: Ensuring a National Policy Framework for Artificial Intelligence

Signed December 11, 2025 and published in the Federal Register on December 16, 2025, EO 14365 is not aimed at AI companies at all — it is aimed at state governments, specifically the state AI laws covered elsewhere in this cluster. Its operative sections:

• Section 3 (30 days): directs the Attorney General to establish an AI Litigation Task Force dedicated to challenging state AI laws the administration considers inconsistent with federal policy, on constitutional and federal-preemption grounds.
• Section 4 (90 days): directs the Secretary of Commerce to identify state laws that, in the order’s framing, require AI models to alter their truthful outputs or that raise other constitutional concerns.
• Section 5 (90 days): directs the Secretary of Commerce and federal agencies broadly to condition certain federal funding — including Broadband Equity, Access, and Deployment (BEAD) program funds — on states not maintaining what the order calls “onerous” AI laws, and to assess other discretionary grant programs for similar conditions.
• Section 6 (90 days): directs the FCC Chairman to initiate a proceeding on a federal AI disclosure standard intended to preempt state disclosure requirements.
• Section 7 (90 days): directs the FTC Chairman to issue a policy statement on how the FTC’s existing deceptive-practices authority preempts state laws that require AI systems to alter their truthful outputs.
• Section 8: directs the Special Advisor for AI and Crypto and the APST to prepare a proposed legislative framework for Congress that would preempt conflicting state AI laws, with carve-outs described for child safety and government procurement.

EO 14365 does not itself create a pre-deployment testing duty, an evaluation duty, or an incident-reporting duty for AI developers — there is no section of the order that reaches a company directly. What it does is target the layer of AI regulation that already has those duties: state law.

How the two EOs relate to the state-law layer

This is the piece that makes the two orders worth reading together rather than separately. The specific duties that frontier AI developers are actually subject to right now — publishing a safety framework, running pre-deployment evaluations, reporting a safety incident within a fixed deadline — come almost entirely from the state layer this cluster covers, not from federal executive action:

• California SB 53 requires large frontier developers to publish a frontier AI framework and transparency reports, and to report a critical safety incident within 15 days.
• New York’s RAISE Act imposes a comparable frontier-model safety and incident-disclosure regime.
• Colorado’s AI Act and Utah’s AI policy law impose their own consumer-protection-oriented AI disclosure duties.
• Texas TRAIGA adds a fourth, differently-scoped state approach.

EO 14179 removed the one federal reporting duty that used to sit alongside these state laws (EO 14110’s DPA-based red-team reporting requirement). EO 14365, ten months later, goes further and directly targets the state laws themselves — using DOJ litigation, a federal disclosure standard meant to preempt state disclosure rules, an FTC preemption theory, and conditional federal funding — because, as of December 2025, those state laws are the only layer of U.S. AI regulation still imposing binding evaluation and incident-reporting duties on developers. Whether any of that preemption effort succeeds is a live, unresolved legal question turning on ordinary constitutional preemption doctrine; nothing in EO 14365 itself invalidates a state law by its own force. Until a court, an act of Congress, or a state repeals its own statute changes that, the state laws above remain in effect exactly as written.

How this compares to the EU AI Act

The EU AI Act takes a structurally different approach than either U.S. layer: a single, comprehensive, risk-tiered regulation, adopted by the EU’s ordinary legislative process, that applies uniformly across all 27 member states with its own enforcement body (see CASRAI’s jurisdiction map of AI regulation worldwide and the EU AI Act high-risk compliance checklist). Neither EO 14179 nor EO 14365 resembles that model. The U.S. federal layer, as these two orders currently stand, does not set substantive AI safety requirements at all — it sets federal executive-branch policy and, in EO 14365’s case, actively works to prevent state-level rules from filling that gap. A company operating in both the EU and the U.S. today faces one binding comprehensive federal-style framework (the EU AI Act) and a patchwork of state laws in the U.S., with federal executive action currently pointed at shrinking the U.S. patchwork rather than replacing it with a federal equivalent.

Pre-EO context: the AI Bill of Rights blueprint

Before either the Biden or Trump AI executive orders, the White House Office of Science and Technology Policy released the Blueprint for an AI Bill of Rights in October 2022 — a non-binding white paper, not an executive order, setting out five principles for automated systems. It was explicitly non-binding when written and was never enforceable; EO 14179’s revocation of EO 14110 does not touch it directly, since the Blueprint was never itself the subject of an executive order. It is useful context for how far federal AI policy has moved: from a voluntary five-principle statement in 2022, to a binding federal reporting requirement under EO 14110 in 2023, to that requirement’s repeal under EO 14179 in 2025, to an executive effort to preempt state-level requirements under EO 14365 by the end of that same year.

The terminology these orders touch, and how CASRAI tracks it

EO 14110’s revoked reporting requirement centered on AI red-team testing and evaluation results for the largest models — concepts that CASRAI’s own NIKOLAI dictionary catalogs in its N5 — Evidence and evaluations track, covering elements like evaluation, evaluation run, elicitation method, and saturation status. That is the specific vocabulary a federal reporting requirement in this space would need to define, and it is the same vocabulary the labs’ own safety frameworks already use, inconsistently, across their public documents.

Similarly, the incident-reporting duties EO 14365 is trying to preempt at the state level — California SB 53’s 15-day critical safety incident reporting window, New York’s RAISE Act equivalent — are exactly the concepts NIKOLAI’s N7 — Incidents track maps: incident, incident type, discovery method, and the reporting deadline and recipient a given law or framework specifies. CASRAI tracks this terminology because state law, lab safety frameworks, and (until its 2025 revocation) federal executive action have each used their own words for overlapping ideas, and NIKOLAI is CASRAI’s own attempt at an index across them — not a standard, and not endorsed by any regulator, lab, or evaluator that appears in its crosswalks. See CASRAI’s own explainer on what NIKOLAI is for the full picture, including how its Mapping Declarations process lets an organization confirm or dispute how NIKOLAI has mapped its own framework.

Frequently asked questions

What is Trump’s AI executive order called?

There are two, not one. EO 14179, “Removing Barriers to American Leadership in Artificial Intelligence,” signed January 23, 2025, revoked the prior administration’s AI executive order and set up the July 2025 AI Action Plan. EO 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” signed December 11, 2025, directs federal agencies to challenge and work to preempt state AI laws.

Does EO 14179 or EO 14365 regulate AI companies directly?

No. Both orders direct federal agencies — what to review, what guidance to revise, what plan to write, what legal theories to develop, what funding conditions to attach. Neither order, by itself, creates a new legal duty an AI developer must comply with. EO 14179’s main effect on companies was removing a federal reporting duty that existed under the order it revoked; EO 14365’s is an attempt to remove state-level duties, not to add a federal one.

What happened to Biden’s AI executive order?

Executive Order 14110, “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,” signed October 30, 2023, was revoked in full by Section 3 of EO 14179 on January 23, 2025. Its Defense Production Act-based reporting requirement for the largest dual-use foundation models ended with that revocation.

Do federal AI executive orders override state AI laws like California’s SB 53?

Not automatically, and not yet. An executive order cannot preempt a state law by itself — EO 14365 directs agencies to pursue preemption through litigation, a federal disclosure standard, an FTC policy statement, and proposed legislation, but none of those mechanisms has, on its own authority as an executive order, invalidated a state law. Until a court rules, Congress acts, or a state changes its own statute, laws like California SB 53, the New York RAISE Act, Colorado’s AI Act, Utah’s AI policy law, and Texas TRAIGA remain in effect.

Is there a federal AI incident reporting or pre-deployment testing requirement right now?

Not from either Trump AI executive order. The one federal requirement in this area, created by the now-revoked EO 14110, required Defense Production Act reporting of red-team safety testing results for the largest dual-use foundation models; EO 14179 ended it. The binding incident-reporting and evaluation requirements that currently exist in the U.S. are at the state level, which is exactly what EO 14365 is aimed at.

How does U.S. federal AI policy compare to the EU AI Act?

Differently in kind, not just degree. The EU AI Act is a single comprehensive, risk-tiered statute applying across the EU. Neither U.S. executive order sets comparable substantive requirements — EO 14179 set internal federal policy and removed an existing reporting duty, and EO 14365 works to narrow the state laws that currently supply the closest U.S. equivalent to binding, EU-AI-Act-style obligations.