🏛️ Issues and Advocacy Training
Courses › Energy, AI and Data Center Policy: What the Orders Say

Lesson 3 of 9 · about 19 minutes

The AI Orders: From Revoking EO 14110 to 'Super Intelligence'

You'll be able to

  • Put the main 2025 and 2026 federal AI actions in order with correct numbers, titles and dates
  • Say which order actually revoked EO 14110 and what EO 14179 added
  • Explain what EO 14319 requires of AI vendors and what it leaves to OMB guidance
  • Compare EO 14409's voluntary frontier model access with EO 14110's mandatory reporting

The lesson AI Law and Regulation covers the EU AI Act, the state laws, and EO 14365's attack on state AI rules. This lesson covers everything else the White House did on AI from January 2025 to October 2026, in order, with what each document actually requires. One correction first, because it is repeated so often.

Who revoked the Biden AI order

President Biden's EO 14110, signed October 30, 2023, was the most detailed federal AI order to date. Among other things it used the Defense Production Act to require developers of large "dual-use foundation models" to report safety test results to the government.

It was revoked on January 20, 2025, by EO 14148, "Initial Rescissions of Harmful Executive Orders and Actions," a single order that cancelled dozens of Biden orders at once. EO 14179, signed three days later and titled "Removing Barriers to American Leadership in Artificial Intelligence," refers to "the revoked Executive Order 14110." Its job was cleanup and direction: review everything agencies had done under 14110 and suspend or rescind what conflicts with the new policy, and produce an AI action plan within 180 days. Its policy line is to "sustain and enhance America's global AI dominance." Many summaries, including some law firm alerts, say EO 14179 revoked 14110. Even the purpose section of EO 14365 credits EO 14179 with that revocation. But the revocation list itself, item (ggg), is in EO 14148.

The timeline

DateActionWhat the text does
Jan 20, 2025EO 14148Revokes EO 14110 among many others
Jan 23, 2025EO 14179Reviews 14110 actions; orders an action plan in 180 days
Apr 3, 2025OMB M-25-21 and M-25-22Rules for agency AI use and AI purchasing
Jul 23, 2025America's AI Action PlanRecommended actions under three pillars
Jul 23, 2025EO 14318, 14319, 14320Data center permitting (lesson 4); "woke AI" procurement; AI export packages
Nov 24, 2025EO 14363, Genesis MissionDOE platform to apply AI to science
Dec 11, 2025EO 14365 and OMB M-26-04Campaign against state AI laws; guidance on Unbiased AI Principles
Mar 20, 2026National Policy Framework for AILegislative recommendations for federal preemption (lesson 7)
Jun 2, 2026EO 14409Cyber defense and voluntary pre-release access to frontier models
Sep 29, 2026EO 14434Agencies say "Super Intelligence" instead of "AI"

Smaller orders on AI education for young people (EO 14277, April 2025) and AI for pediatric cancer research (EO 14355, September 2025) fit the same pattern of promoting use.

America's AI Action Plan, July 2025

The plan has three pillars: accelerate innovation, build AI infrastructure, and lead in international AI diplomacy and security. It is signed by Michael Kratsios (science adviser), David Sacks (Special Advisor for AI and Crypto) and Marco Rubio (national security adviser). Its slogan for infrastructure is "Build, Baby, Build!"

Some lines matter more than the slogans:

  • States. Federal AI-related funding should not go to states with "burdensome" AI rules, but the government should also not interfere with states' right to pass "prudent laws that are not unduly restrictive." OMB is to have agencies weigh a state's AI regulatory climate in discretionary funding. The FCC is to look at whether state AI rules interfere with its authority. This was the seed of EO 14365.
  • Standards. NIST is to revise its AI Risk Management Framework to remove references to misinformation, diversity, equity and inclusion, and climate change.
  • Open models. The plan encourages open-source and open-weight models.
  • Infrastructure. New NEPA categorical exclusions for data centers, wider use of the FAST-41 permitting process, a possible nationwide Clean Water Act permit, federal land for data centers and power plants, and a grid strategy to "stabilize," "optimize" and then expand.

EO 14319: "Preventing Woke AI in the Federal Government"

The order's purpose section calls diversity, equity and inclusion an "existential threat to reliable AI" and gives examples of image generators changing the race or sex of historical figures. The operative part is narrower than the title. Agencies may procure only large language models built under two "Unbiased AI Principles":

  • Truth-seeking: truthful answers to factual questions, priority on historical accuracy and scientific inquiry, and acknowledging uncertainty.
  • Ideological neutrality: no intentional encoding of partisan or ideological judgments unless the user prompts for them or can readily see them.

The guidance had to account for technical limits, let vendors comply by disclosing a model's system prompt, specifications or evaluations without handing over model weights, avoid over-prescription, and allow exceptions for national security systems. New contracts must include these terms and make the vendor pay decommissioning costs if the agency ends a contract for noncompliance after a chance to cure. The order states that the government "should be hesitant to regulate" AI in the private market. It reaches private companies only as vendors.

Supporters say taxpayers should not buy tools that slant facts. The open questions are practical: who decides what counts as neutral, how a vendor proves a negative, and whether a model tuned to satisfy one administration's idea of neutrality would satisfy the next one's.

Example: checking the OMB deadline

EO 14319 was signed July 23, 2025 and gave OMB 120 days.

  • Days left in July after the 23rd: 8.
  • Through August: 8 + 31 = 39. Through September: 39 + 30 = 69. Through October: 69 + 31 = 100.
  • 20 more days into November: due November 20, 2025.
  • OMB's index lists M-26-04, "Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles," dated December 11, 2025. November 21 to 30 is 10 days, plus 11 days of December: 21 days late.

Late but delivered. Agencies then had 90 days to adopt procedures, which puts that deadline around March 11, 2026.

Exports and science: EO 14320 and EO 14363

EO 14320 tells Commerce, with State and the science office, to set up an American AI Exports Program within 90 days and to call for proposals from industry consortia offering "full-stack" packages: chips, cloud, models and applications. The stated aim is to have other countries build on U.S. technology instead of a rival's.

EO 14363 creates the Genesis Mission. The Energy Secretary runs it inside DOE, the President's science adviser leads it across agencies, and DOE must build an "American Science and Security Platform" combining national lab computing and data for AI-driven research.

EO 14409: security without licensing

The June 2, 2026 order is the most interesting of the group because it adds government testing while denying it is regulation. Within 60 days, Treasury, the NSA (through the Secretary of War) and CISA were to build a classified benchmark of advanced cyber capabilities and set a threshold for a "covered frontier model," with the NSA Director making the call. They were also to design a voluntary framework for developers to give the government access to covered models for up to 30 days before releasing them to other trusted partners. Other sections order fast cyber defense of federal systems, an AI cybersecurity clearinghouse under Treasury, and priority prosecution of people who use AI to break into computers.

EO 14110 (Biden, 2023)EO 14409 (Trump, 2026)
TriggerLarge dual-use foundation modelsModels past a classified cyber capability threshold
DutyReport safety test results, under the Defense Production ActVoluntary pre-release access up to 30 days
Lead agenciesCommerceNSA, CISA, Treasury
StatusRevoked January 20, 2025In effect; no public framework document found in this review, October 2026

A fair reading: both administrations decided the government should see the most capable models early. They disagree about whether that should be required.

A second testing track runs through Commerce, and it came first. On May 5, 2026, a month before EO 14409, CIO Dive reported that NIST's Center for AI Standards and Innovation (CAISI) had signed agreements with Google DeepMind, Microsoft and xAI for pre-deployment evaluations and research on their frontier models. The government can test a model before public release and keep assessing it after. The agreements build on the 2024 deals NIST's predecessor office, the U.S. AI Safety Institute, signed with OpenAI and Anthropic, with terms updated to follow CAISI and Commerce Secretary directives and the AI Action Plan. CAISI Director Chris Fall called independent measurement "essential to understanding frontier AI." These are agreements, not rules. Nothing in the reporting says a company that declines faces a penalty. One analyst quoted by CIO Dive, Nick Patience of The Futurum Group, argued the pull is commercial anyway: companies buying AI for federal work will prefer vendors in good standing with Commerce and NIST. NIST's announcement page, linked from the article, would not load for this review, so the details here rest on CIO Dive's account.

TrackWho testsCompanies namedBinding?
CAISI agreements (2024, expanded May 2026)NIST, Commerce DepartmentOpenAI, Anthropic (2024); Google DeepMind, Microsoft, xAI (2026)No; signed agreements
EO 14409 framework (June 2026)NSA, with CISA and TreasuryNone public as of October 2026No; voluntary access up to 30 days

Common mix-ups

  • Claim: EO 14179 revoked Biden's AI order. What the evidence shows: EO 14148 did, on January 20, 2025. EO 14179 ordered the cleanup and the action plan.
  • Claim: The "woke AI" order regulates ChatGPT and other consumer chatbots. What the evidence shows: It sets terms for what federal agencies buy. A company that does not sell to the government is not covered.
  • Claim: The administration has no AI safety testing. What the evidence shows: EO 14409 sets up voluntary pre-release access and classified cyber benchmarking through the NSA. It is voluntary, and no public framework document turned up in this review.
  • Claim: EO 14434 declared that superintelligence exists. What the evidence shows: It changes the word agencies use and defines "SI" by the existing statutory definition of AI.

Reporting angle

  • Ask federal agency offices that serve your area (USDA, VA, SSA, BLM, Forest Service) which AI tools they use, whether those tools were bought under M-26-04 terms, and where their AI use case inventory is posted.
  • Ask your state's AI and IT office whether federal grants it receives now carry conditions about state AI law, as the Action Plan and EO 14365 propose.
  • Track the EO 14409 framework. Ask AI developers whether they have joined it and what the NSA threshold is. Expect classified answers; ask anyway and report the refusal.
  • Ask NIST's CAISI how many frontier models it has tested under its agreements, whether any release was delayed or changed because of a test, and whether results go to Congress. Request the signed agreements under FOIA (5 U.S.C. 552) if they are not posted.
  • For any AI claim in a political speech, find the order number. If there is none, it is a talking point.

Research angle

Check yourself

1. Which executive order revoked President Biden's AI order, EO 14110?

2. Under EO 14319, 'Preventing Woke AI in the Federal Government,' what must agencies do?

3. EO 14319 gave OMB 120 days from July 23, 2025 to issue guidance. OMB issued M-26-04 on December 11, 2025. How late was it?

4. What does EO 14409 (June 2, 2026) set up for frontier AI models?

5. Who signed the introduction to America's AI Action Plan (July 2025)?

Terms in this lesson

Dual-use foundation model
Term in Biden's EO 14110 for large general models that could pose security risks; developers had to report safety test results to the government under the Defense Production Act.
Covered frontier model
Term in EO 14409 for an AI model whose advanced cyber capabilities pass a threshold set through a classified benchmark run with the NSA.
Unbiased AI Principles
EO 14319's two procurement standards for large language models: truth-seeking and ideological neutrality.
Open-weight model
An AI model whose trained parameters are released so anyone can run or modify it. The AI Action Plan encourages them.
Genesis Mission
A DOE-led effort created by EO 14363 (November 2025) to apply AI to scientific discovery using national lab data and computing on a shared platform.

Full glossary

Sources