Executive Order 14110 (RESCINDED)

Executive Order 14110, signed by President Joseph R. Biden, Jr. on October 30, 2023, was an omnibus AI executive order directing more than 100 agency actions across safety, civil rights, privacy, labor, immigration, healthcare, competition, and international engagement. Its provisions included dual-use foundation model reporting at a 10²⁶ FLOPs compute threshold, NIST safety standards, Infrastructure-as-a-Service know-your-customer requirements, content-provenance standards, biosecurity measures, and federal-use governance. It was rescinded by Trump's EO 14148 on January 20, 2025.

Superseded / Rescinded: Rescinded by Trump EO 14148 ("Initial Rescissions of Harmful Executive Orders and Actions," signed January 20, 2025) (Source: federalregister.gov). The successor framework is Executive Order 14365 — Ensuring a National Policy Framework for AI (federal preemption posture) plus America's AI Action Plan. Kept for historical reference because downstream work (NIST AI 600-1, US AI Safety Institute, BIS compute-reporting, agency Chief AI Officers) was commissioned under this order and partially persists.

Scope

The order was the longest AI-focused executive order issued to that date, spanning more than 100 agency actions across safety, civil rights, privacy, labor, immigration, healthcare, competition, and international engagement. As published in the Federal Register, its stated principles included prioritizing safety and security in AI systems, ensuring equity and civil rights, protecting consumers and privacy, and supporting workers through technological transitions; it also established a White House Artificial Intelligence Council to coordinate implementation across agencies (Source: federalregister.gov).

Key provisions

The order's central regulatory mechanism was dual-use foundation model reporting under § 4.2. Acting under the Defense Production Act, Commerce required developers training models above a compute threshold — operationalized at 10²⁶ FLOPs for dense models, with a lower threshold for biology sequence models — to report training-run plans, the physical and cybersecurity protections in place for model weights, ownership and possession of those weights, and red-team and safety-test results.

Several provisions directed standards and institution-building. Under § 4.1, NIST was directed to develop evaluations for dual-use foundation models, red-teaming guidance, and a generative-AI companion to the AI Risk Management Framework, later published as NIST AI 600-1. The US AI Safety Institute was established within NIST as the US counterpart to the UK AISI; it survived the rescission and was restructured under the Trump administration.

The order also addressed infrastructure access, media authenticity, and biosecurity. Under § 4.2(c), US Infrastructure-as-a-Service providers were required to collect and report identifying information on foreign customers training large AI models. Under § 4.5, Commerce was directed to develop standards for watermarking and content-authentication of AI-generated media. Under § 4.4, HHS was directed to address AI-enabled CBRN risks, including screening of synthetic nucleic-acid providers.

On civil rights, § 7 directed agency guidance on AI-enabled discrimination in housing, lending, employment, healthcare, and criminal justice, building on the Blueprint for an AI Bill of Rights and the AI RMF.

For the federal government's own use of AI, § 10 directed a Chief AI Officer at every agency, an inventory of federal AI uses, and generative-AI pilots; the agency CAIO structure was retained under Executive Order 14365 — Ensuring a National Policy Framework for AI. On immigration, § 5 streamlined visas for AI talent and expanded Schedule A occupations. On international engagement, § 11 directed common approaches with allies and contributed to the The Bletchley Declaration (AI Safety Summit, 1–2 November 2023) and the G7 Hiroshima process.

Key definitions

• Dual-use foundation model — a model trained on broad data, self-supervised, with at least tens of billions of parameters, that possesses or is easily modifiable to possess high-risk capabilities (CBRN uplift, offensive cyber, evasion of human control).
• AI red-teaming — structured adversarial testing of AI systems.
• Generative AI — models that emulate the structure of their input to generate synthetic content.

Comparison with other frameworks

The order's use of a compute threshold as a proxy for frontier capability became a recurring point of contention. EO 14110 operationalized 10²⁶ FLOPs; California SB 1047 — Safe and Secure Innovation for Frontier AI Models Act (enrolled + veto) adopted the same threshold; Governor Newsom vetoed SB 1047 partly on compute-threshold grounds; and California SB 53 — Transparency in Frontier AI Act subsequently moved to revenue-based triggers. The order required reporting but not licensing, a less stringent posture than the EU AI Act (Regulation 2024/1689), which went further, and than SB 1047, which would have added liability. Its standards instruments differed in their binding status: the NIST AI Risk Management Framework (AI RMF 1.0) and NIST AI 600-1 — Generative AI Profile were and remain voluntary, while EO 14110 layered mandatory reporting on top of them. On federalism, Executive Order 14365 — Ensuring a National Policy Framework for AI reversed the order's approach of nudging state laws to align with federal policy, instead preempting stricter state laws.

Relationships

• supersedes: Executive Order 13859 — Maintaining American Leadership in Artificial Intelligence — Trump-era 2019 AI EO (conceptually superseded)
• superseded-by: Executive Order 14365 — Ensuring a National Policy Framework for AI — rescinded via EO 14148 and framework reversed
• depends-on: NIST AI Risk Management Framework (AI RMF 1.0)
• supports: NIST AI 600-1 — Generative AI Profile, BIS Framework for AI Diffusion — Interim Final Rule (Jan 13, 2025) — downstream work commissioned under this order
• related: California SB 1047 — Safe and Secure Innovation for Frontier AI Models Act (enrolled + veto) — adopted the same compute threshold
• contradicts: America's AI Action Plan, Executive Order 14365 — Ensuring a National Policy Framework for AI
• related: Statement on AI Risk (CAIS), FLI — Pause Giant AI Experiments: An Open Letter — advocacy context

Sources

• EO 14110 source summary