United States — AI policy

The market-driven pole of the wiki's governance map — and the clearest case of a major power declining a horizontal AI law. The US relies on sectoral + state-level rules and executive action that has swung deregulatory, in deliberate contrast to the eu-ai-act's rights-driven horizontal regime and China's state-driven one. The sectoral route is not one position, though: the UK reaches it from the opposite direction, issuing cross-sector principles centrally and delegating enforcement to existing regulators, where the US delegates the rule-making itself to sectors and states.

The arc

• Biden Executive Order (Oct 2023) — required developers of the most capable models to share safety-test results with the government before release. Subsequently rescinded.

• Trump deregulatory turn — EO 14365 (12 Dec 2025), "Eliminating State Law Obstruction of National Artificial Intelligence Policy." Named and detailed on re-read 2026-08-30; this entry had carried no order number. It directs agencies toward a "unified national approach" and to challenge state AI laws in court — a federal effort aimed not at regulating AI but at clearing regulation away. Three mechanisms it actually uses:

  • a DOJ AI Litigation Task Force, operative from 10 January 2026, to sue over state AI laws as burdens on interstate commerce or as preempted;
  • conditioning $42 billion of already-allocated BEAD broadband money on states repealing AI rules deemed onerous — the money, not the courtroom, being the sharper instrument;
  • carve-outs from preemption for child-safety protections, AI compute and data-centre infrastructure, and state procurement and use of AI.

  And it has not displaced anything. Preemption normally flows from an act of Congress, not an executive order, and Congress has passed no federal AI law. Commentary through mid-2026 reads existing state statutes as largely unaffected in the near term. So the deregulatory pole of this page is, so far, a live push rather than settled law — which matters, because the page's four-way instrument comparison treats it as the US position.

• Sectoral + state patchwork — no federal horizontal law; finance/health/transport handle AI under existing sector rules, while California and Colorado pass piecemeal AI statutes — the fragmentation the Trump EO seeks to preempt.

• Security-lever regime (June 2026) — and the lever is softer than this page said. Cutting against the deregulatory arc, the administration issued Executive Order 14409 (2 June 2026). This entry described it as a gate that "requires submitting new models 30 days before release." Corrected 2026-08-30 against the Congressional Research Service's explainer (T1): the review is voluntary — companies are asked to participate — and the order "stops short of formal licensing or requiring preclearance." It does not even define "covered frontier model"; that was a deliverable due 1 August 2026 out of a classified benchmarking process.

  What remains true is the sequence: within a fortnight, the Anthropic Fable 5 / Mythos 5 suspension (worldwide, over a contested jailbreak claim) and the GPT-5.6 access restriction (release limited to government-approved partners, approved customer-by-customer). CRS lists the Anthropic action among the administration's other approaches, set in contrast to the EO's voluntary assessment — so on the best-sourced reading these were not enforcement of EO 14409 but separate exercises of other authority. Both readings are held on eo-14409.

  The reframing sharpens rather than dissolves the tension this bullet exists for. The administration deregulates domestic conduct by order while restricting specific frontier releases by discretion, and the standing instrument it wrote for the second is one nobody has to obey. The US cell of the wiki's comparison is therefore two ad-hoc actions and a voluntary framework, not a regime.

What is not there (re-read 2026-08-30)

This page had gone two months without a read and the queue flagged it as the one ai-governance page that genuinely ages. The re-read found two errors of characterization above and one absence worth stating: no federal AI statute exists. Everything on this page is executive action, litigation posture, or state law. Two bills would change that — the Artificial Intelligence Civil Rights Act of 2025 (H.R. 6356), which would mandate pre-deployment evaluation of certain models, and a discussion draft of the Great American Artificial Intelligence Act of 2026 — and neither has passed (crs-eo-14409-explained).

One more thing the corpus lacked: NIST already held pre-release access to Anthropic's and OpenAI's major models by private agreement, before any of these orders, with no public transparency and no coverage of other labs. The practice of government pre-release model access in the US began as a commercial arrangement, not a policy.

Where it sits on the wiki's axes

• The fourth model. With the US paged, the spoke's governance map is a clean four-way comparison of instrument philosophy: EU rights-driven horizontal-binding · China state-driven content-binding · US market-driven fragmented/deregulatory · Israel innovation-driven sectoral-soft (responsible-innovation). The US and Israel share the light-touch instinct, but for different reasons (preemptive deregulation vs. deliberate hub-strategy).
• Federal-vs-subnational tension. The US adds an axis the others lack: an internal preemption fight (federal deregulation vs. state statutes) — governance contested within a jurisdiction, not just across them.
• Convergence caveat holds. Even amid divergence on instruments, the US still references the oecd-ai-principles definitional layer the synthesis identifies as the shared substrate.

Related

eu-ai-act · china-ai-regulation · responsible-innovation · risk-based-regulation · nist-ai-rmf · export-controls-on-ai · eo-14409 · anthropic-export-ban-2026 · openai-gpt56-access-restriction · uk-ai-regulation · ai-governance