Summary of Differences Between Draft and Final AI Executive Order

Here’s the cleanest “redline-style” comparison between the draft Executive Order the final EO posted on the White House site today. (The White House)

This summary focusing on substantive text changes.

1. Title / Framing

Caption in the draft PDF (under “EXECUTIVE ORDER”):

“ELIMINATING STATE LAW OBSTRUCTION OF NATIONAL AI POLICY”

Public title on WhiteHouse.gov:

“ENSURING A NATIONAL POLICY FRAMEWORK FOR ARTIFICIAL INTELLIGENCE”

Effect: The public-facing title is softened from “eliminating obstruction” to “ensuring a national policy framework,” but the underlying preemption project is intact.

2. Section 1 – Purpose

Structure and overall thrust are the same, but the rhetoric is noticeably toned down.

2.1. “American” → “United States”

Draft:

“…will promote American national and economic security…”
Final:
“…will promote United States national and economic security…”

Same in the later reference to “American AI leadership” → “United States AI leadership.”

2.2. Deleted: California SB 53 + “significant catastrophic risk”

The draft explicitly calls out California’s new AI law (SB 53) as:

a “complex and burdensome disclosure and reporting law”
“premised on the purely speculative suspicion that AI might ‘pose significant catastrophic risk’.”

All of that California-specific language is gone from the final EO. The published version only keeps the Colorado example.

2.3. Rewritten “patchwork / fear-based” language

Draft Purpose includes a much sharper attack on state laws. In substance, it says:

State legislatures have introduced “over 1,000 AI bills that threaten to undermine” an “innovative culture.”
Colorado’s “algorithmic discrimination” law may force models to embed DEI and “produce false results” to avoid “differential treatment or impact.”
“Sophisticated proponents of a fear-based regulatory capture strategy are responsible for inciting these laws…”
This creates a “patchwork” that forces compliance with the “lowest common denominator” and lets restrictive states “dictate national AI policy.”

In the final EO, that is replaced with a more structured, three-part formulation:

Patchwork: “State-by-State regulation by definition creates a patchwork of 50 different regulatory regimes… particularly for start-ups.”
Ideology: “State laws are increasingly responsible for requiring entities to embed ideological bias within models.”
Extraterritoriality: “State laws sometimes impermissibly regulate beyond State borders, impinging on interstate commerce.”

The Colorado example survives but is softened:

Draft: explicitly “embed DEI in their programming” and “produce false results” for Colorado’s enumerated demographic groups.
Final: no “DEI” reference; just that the law may “force AI models to produce false results” to avoid differential impact on “protected groups.”

2.4. Added: Positive checklist for a national framework

The final EO keeps the “minimally burdensome national standard” idea, but adds an explicit checklist: the framework should ensure:

children are protected,
censorship is prevented,
copyrights are respected, and
communities are safeguarded.

That explicit list is not spelled out this way in the draft.

3. Section 2 – Policy

Very small but real change.

Draft Sec. 2 (substance):

“…through a minimally burdensome, uniform national policy framework for AI.”

Final Sec. 2:

“…through a minimally burdensome national policy framework for AI.” (the word “uniform” is dropped).

So the policy still clearly implies uniformity, but the term itself is removed.

4. Section 3 – AI Litigation Task Force

Core idea is the same, but the scope is narrowed and clarified.

4.1. Narrowing the target

Draft: AI Litigation Task Force’s “sole responsibility” is to challenge State AI laws, on grounds they unconstitutionally regulate interstate commerce, are preempted, or otherwise unlawful.

Final adds an important qualifier:

“…whose sole responsibility shall be to challenge State AI laws inconsistent with the policy set forth in section 2 of this order…”

That limiting clause (“inconsistent with the policy set forth in section 2”) is new in the final EO.

4.2. Defined term

Final also formally defines the shorthand:

“AI Litigation Task Force (Task Force)…” – the parenthetical wasn’t there in the draft.

4.3. Consultative list tidied

The consult list is basically the same (Special Advisor for AI and Crypto; AP for Science and Technology; AP for Economic Policy; Counsel), but the final text regularizes the phrasing (e.g. “the Assistant to the President and Counsel to the President”) and drops draft’s page-break artifacts.

5. Section 4 – Evaluation of State AI Laws

5.1. Heading softened

Draft heading:

“Evaluation of Onerous State AI Laws.”

Final heading:

“Evaluation of State AI Laws.”(The White House)

“Onerous” is removed from the title, though it still appears in the body.

5.2. Authority citation trimmed

Draft: Commerce’s review is “consistent with his authorities under 47 U.S.C. 902(b) and 15 U.S.C. 272(b)” (NTIA + NIST).

Final: “consistent with the Secretary’s authorities under 47 U.S.C. 902(b)” – the explicit NIST cite (15 U.S.C. 272(b)) is dropped.

5.3. New pro-state clause

The final EO adds a sentence at the end of Sec. 4:

“The evaluation may additionally identify State laws that promote AI innovation consistent with the policy set forth in section 2 of this order.”

That “good laws” clause does not appear in the draft.

Substantive core (identifying laws that require altering truthful outputs or compel speech in ways that may violate the First Amendment) appears in both.

6. Section 5 – Restrictions on State Funding

BEAD hook survives with tweaks and a bit of narrowing.

6.1. BEAD cite and “Benefit of the Bargain”

Both versions tie BEAD non-deployment funds to state AI laws and reference “Benefit of the Bargain” reforms and 47 U.S.C. 1702(e)-(f). The final simply cleans up the statute cite (OCR in the draft made the (f) messy) and smart quotes.(New Jersey League of Municipalities)

6.2. “Onerous AI laws” clarification

Final adds “onerous” in the operative clause:

Draft: States with AI laws identified under Sec. 4 are ineligible for non-deployment funds.
Final: States with onerous AI laws identified under Sec. 4 are ineligible.

That makes the funding penalty text track the “onerous” concept more tightly.

6.3. Discretionary grants – performance period language

In Sec. 5(b), the structure is the same (agencies are to condition discretionary grants on states not enacting or not enforcing conflicting AI laws), but:

Draft: refers more loosely to not enforcing such laws while receiving funding.
Final: specifies “during the performance period in which it receives the discretionary funding.”

That’s a slightly more technical, grants-law-friendly formulation.

7. Section 6 – Federal Reporting and Disclosure Standard

This section is substantively unchanged.

Both:

Direct the FCC Chair, in consultation with the Special Advisor for AI and Crypto, to consider adopting a Federal reporting and disclosure standard for AI models that preempts conflicting State laws.

No meaningful policy edits beyond punctuation/spacing.

8. Section 7 – FTC / Deceptive Conduct

Again, substance is the same; language is formalized.

Draft: refers to “the FTC Act’s prohibition on unfair and deceptive acts or practices…”

Final: spells out “the Federal Trade Commission Act’s prohibition on unfair and deceptive acts or practices under 15 U.S.C. 45…”

The operative idea—FTC policy statement explaining when state laws that force non-truthful AI outputs are preempted as “deceptive”—is unchanged.

9. Section 8 – Legislation

This is the other big change after Section 1.

9.1. Who drafts the bill

Draft: Section 8 tasks the Special Advisor for AI and Crypto plus a more “legislative-shop” person (Director of the Office of Legislative Affairs / equivalent) with preparing the preemptive federal framework.

Final:

“The Special Advisor for AI and Crypto and the Assistant to the President for Science and Technology shall jointly prepare a legislative recommendation…”

So the science/technology office is explicitly brought in; the Legislative Affairs-centric reference from the draft is removed.

9.2. Carve-outs

Both versions say the legislative recommendation should not propose preempting certain types of state laws, but the final version tightens and expands the carve-outs:

Final Sec. 8(b) says the recommendation shall not propose preempting otherwise lawful State AI laws relating to:

“child safety protections;”
“AI compute and data center infrastructure, other than generally applicable permitting reforms;”
“State government procurement and use of AI;” and
“other topics as shall be determined.”

Changes vs. draft:

“other than generally applicable permitting reforms” is a new limiting clause – it clearly allows preemption of state attempts to use permitting as a back-door AI regulation, while preserving generic infrastructure permitting.
The catch-all “other topics as shall be determined” appears only in the final EO, giving more flexibility on future exclusions.

10. Section 9 – General Provisions / Signature

Text of Sec. 9(a)–(d) is effectively the same standard boilerplate (no change to the “no private right of action” clause). The differences are:

Draft: no signature block (it’s a deliberative draft).

Final: adds the signature and dating lines:

“DONALD J. TRUMP
THE WHITE HOUSE,
December 11, 2025.”

Quick “Redline” Summary by Theme

Tone and targeting in Purpose
- Deleted: California-specific SB 53 language and “significant catastrophic risk” rhetoric; “fear-based regulatory capture strategy”; “lowest common denominator” phrase.
- Softened: Colorado / DEI passage (still attacks the law, but removes “embed DEI” language).
- Added: Positive checklist (protect kids, prevent censorship, respect copyrights, safeguard communities).
Scope of federal pushback
- Narrowed: AI Litigation Task Force now limited to state laws “inconsistent with the policy set forth in section 2.”
- Softened branding: Sec. 4 heading no longer says “Onerous” in the title, but “onerous” remains in the operative clause and in BEAD-funding penalty text.
Federalism gestures
- Added: Sec. 4 now explicitly says Commerce may identify state laws that promote AI innovation.
- Carve-outs refined: Sec. 8(b) keeps carve-outs for child safety, data centers, and procurement but adds the permitting qualifier and an “other topics” catch-all.
Technical / legal clean-up
- “American” → “United States” throughout.
- “FTC Act” → “Federal Trade Commission Act”.
- Small shifts in who drafts the legislative package (Leg Affairs → AP for Science & Tech).
- More precise grant-funding language (“performance period”).
- Removal of an explicit NIST statutory cite in Sec. 4.