A History of US Federal AI Executive Orders

Six years, three administrations, eight executive orders, and not one of them required Congress to do anything. That's the actual shape of US federal AI policy since 2019 — not a law you can look up in the US Code, but a sequence of presidential directives that has already reversed itself once, in two separate steps three days apart. If your compliance program treats "the AI executive order" as a single, stable reference point, it's tracking the wrong kind of document.

The record, in order

Here's the full sequence before the detail, so you can place any single order in context:

• February 11, 2019 — EO 13859, "Maintaining American Leadership in Artificial Intelligence" (Trump), launching the American AI Initiative.
• December 3, 2020 — EO 13960, "Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government" (Trump), governing agencies' own AI use.
• October 30, 2023 — EO 14110, "Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence" (Biden), the most comprehensive order to date.
• January 20, 2025 — EO 14148, "Initial Rescissions of Harmful Executive Orders and Actions" (Trump), formally revoking EO 14110 along with dozens of other Biden-era orders.
• January 23, 2025 — EO 14179, "Removing Barriers to American Leadership in Artificial Intelligence" (Trump), setting the actual replacement policy and ordering a 180-day AI Action Plan.
• July 23, 2025 — the AI Action Plan itself, "Winning the Race," released alongside three more orders: accelerated data-center permitting, promotion of AI technology exports, and EO 14319, "Preventing Woke AI in the Federal Government."

Two things worth noticing before the deep dive: the pace has been accelerating, not slowing down, and every single item on that list came from the Oval Office, not from Capitol Hill.

Why there's no Obama-era AI executive order

It's a common assumption that federal AI policy started earlier than 2019 — that some Obama-era order laid the groundwork before Trump's first term. It didn't happen that way. What the Obama administration actually produced, in October 2016, was an Office of Science and Technology Policy report, "Preparing for the Future of Artificial Intelligence," paired with a National AI R&D Strategic Plan. Both were substantive, and both are still cited in later federal AI documents. Neither was an executive order — a distinction that isn't pedantic. An order directs agency action under the president's own authority; a strategy report from an advisory office is a recommendation with no directive force behind it. The gap between the 2016 report and 2019's EO 13859 is the gap between "here's what the government should consider" and "here's what agencies are now required to do."

2019: EO 13859 and the American AI Initiative

EO 13859, signed February 11, 2019, established the American AI Initiative and framed the entire order around a single goal: keeping the US ahead of other countries in AI research and deployment. It's coordinated through the National Science and Technology Council's Select Committee on Artificial Intelligence, and it directs agencies to prioritize AI in their R&D budgets, expand researcher access to federal data and computing resources, and begin work on technical standards for trustworthy AI systems.

What it isn't is a risk-management order — no safety-testing requirement, no civil-rights directive. The frame throughout is economic and strategic competitiveness, not harm mitigation, and that framing matters: it's the same frame EO 14179 would return to six years later, after a detour through a very different kind of order in between.

2020: EO 13960 turns inward, toward the government's own AI use

EO 13960, signed December 3, 2020, is a narrower and more operational document. Where EO 13859 pointed outward — American AI competitiveness generally — EO 13960 points inward, at how federal agencies use AI themselves, outside national-security and defense contexts. It sets nine principles federal AI use is supposed to meet: lawful, respectful of American values, purposeful and performance-driven, accurate and reliable, safe and secure, understandable, responsible and traceable, regularly monitored, and transparent and accountable.

The order also requires agencies to inventory their own AI use cases and check them against those nine principles. That inventory requirement becomes the template every later order reuses — EO 14110's implementing guidance and even 2025's replacement guidance both keep the "maintain an AI use-case inventory" mechanic, even as the substance around it changed twice.

2023: EO 14110, the comprehensive one

EO 14110, signed October 30, 2023, is a different order in scale, not just degree. Where the two Trump-era orders ran a handful of pages each, EO 14110 is genuinely comprehensive: safety-testing and reporting requirements for the most powerful AI models, civil-rights and consumer-protection directives spanning housing, employment, and lending, immigration provisions meant to attract AI talent, and specific direction to NIST and OMB to produce detailed implementing guidance rather than leaving agencies to interpret the order on their own — the piece worth tracing all the way down into what agencies actually did, not just what the order itself said.

What EO 14110 actually produced inside agencies, before it was undone

Here's the part that tends to get lost in headline-level summaries: EO 14110 didn't govern agencies directly so much as it triggered a specific piece of paper that did. On March 28, 2024, OMB issued Memorandum M-24-10, "Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence," implementing the order's direction. M-24-10 required every covered agency to name a Chief AI Officer, publish a public inventory of its AI use cases, and apply minimum risk-management practices — testing, monitoring, human oversight — to any AI system presumed to affect people's rights or safety.

Agencies actually did this. The EEOC, for one, published its own compliance plan for M-24-10, naming a Chief AI Officer and documenting how it would inventory and review its AI systems against the memo's requirements — a real, on-the-record example of an executive order's effects reaching all the way down to a specific agency's internal governance paperwork, not a hypothetical.

Then it reversed. EO 14179 directed OMB to revise M-24-10 within 60 days of the order. OMB did, replacing it outright with M-25-21, "Accelerating Federal Use of AI through Innovation, Governance, and Public Trust," on April 3, 2025 — a little over a year after M-24-10 was issued. The new memo keeps the basic scaffolding (a senior AI official, a use-case inventory) but shifts the emphasis toward faster adoption and lighter review. Same agencies, largely the same underlying AI systems, a materially different governing memo about them within about twelve months. A vendor-compliance checklist built against M-24-10's specific requirements in early 2024 needed a rewrite by mid-2025 — and it changed because the executive order behind it changed, not because Congress amended a law.

January 2025: two orders, three days apart, that reversed course

It's tempting to compress the 2025 reversal into a single event — "Trump signed an order revoking Biden's AI order" — but the actual mechanics ran in two steps. EO 14148, "Initial Rescissions of Harmful Executive Orders and Actions," signed on inauguration day, January 20, 2025, formally rescinded EO 14110 along with roughly sixty other Biden-era orders and memoranda in one sweep. It's a repeal, and only a repeal — it doesn't set out what replaces the rescinded policy on AI specifically.

That affirmative replacement came three days later, on January 23, 2025, with EO 14179, "Removing Barriers to American Leadership in Artificial Intelligence." This is the order that states the new policy: sustain and extend American AI dominance, review and roll back prior actions agencies find inconsistent with that goal, and produce a governmentwide AI Action Plan within 180 days. The two-step structure is also why some summaries describe EO 14110 as rescinded "within hours" of the inauguration and others cite January 23 — both dates are correct, depending on whether you're pointing at the rescission or the replacement policy.

Mid-2025: the AI Action Plan and a second wave of orders

EO 14179's 180-day clock produced its output on July 23, 2025: "Winning the Race: America's AI Action Plan," organized around three pillars — accelerating innovation, building AI infrastructure, and leading in international AI diplomacy and security. The administration paired the plan's release with three more executive orders signed the same day: one accelerating federal permitting for AI data centers, one promoting export of the American AI technology stack to allied countries, and EO 14319, "Preventing Woke AI in the Federal Government."

Of the three, EO 14319 has the clearest direct compliance hook for anyone selling to the federal government: it conditions federal procurement of large language models on "ideological neutrality" criteria OMB was directed to define, meaning a vendor's contract eligibility can turn on how its model handles contested political and social topics, not just its accuracy or security posture.

What the pattern means if you're building a compliance program around this

The practical lesson isn't which order is currently in force — it's what kind of instrument that order is. An executive order isn't a statute. It carries the force of law only as long as the president who signed it, or a like-minded successor, keeps it in place, and the implementing guidance built underneath it (an OMB memo, in every case so far) can be rewritten even faster than the order itself, as the M-24-10-to-M-25-21 turnaround shows. A compliance program anchored entirely to "what the current AI executive order requires" is built on the least durable layer of federal AI policy available.

The more durable layers sit elsewhere. Enacted state statutes don't reverse with a change of federal administration — Colorado's and California's AI laws keep applying regardless of what any president signs. Federal agencies also keep enforcing existing statutes against AI-driven conduct independent of which order is nominally current: the FTC's Section 5 authority over deceptive AI claims didn't depend on EO 14110 and wasn't touched by its rescission, and the same holds for the EEOC's Title VII and ADA authority over algorithmic hiring tools — the EEOC pulled its AI hiring guidance directly because of EO 14179 — and for the CFPB's fair-lending obligations for AI-driven credit decisions, where a separate 2025 guidance-document withdrawal removed the agency's AI-specific circulars without touching the underlying statute either. The habit worth building: track the OMB memo number and the statute an agency actually enforces under, not just whichever executive order is in the news that week. The order signals current priorities; the memo and the statute tell you what you're on the hook for.