AI regulation refers to the laws, executive actions, and international agreements that governments use to govern how artificial intelligence is built and used. As of the mid-2020s, different jurisdictions have taken noticeably different approaches, and there is no single global rulebook. The entries in this library that describe specific policy documents illustrate the main shapes that approach has taken.
The European Union has pursued binding legislation. Regulation (EU) 2024/1689, the AI Act, entered into force on 1 August 2024 and applies a risk-based structure: practices deemed an unacceptable risk are prohibited, high-risk systems must meet defined obligations before reaching the market, some systems carry transparency duties, and the majority face no specific rules. The Act also sets obligations for general-purpose AI models, with rules phasing in on a schedule that runs into 2026.
The United States, by contrast, relied during this period mainly on executive action rather than comprehensive legislation. Executive Order 14110, signed on 30 October 2023, directed federal agencies to develop standards and guidance for safe and trustworthy AI. Because it was an executive order rather than a statute, it could be reversed by a later administration, and it was revoked by Executive Order 14148 on 20 January 2025. This shows how policy resting on executive action can change with a change in administration.
A separate strand is national strategy. China’s State Council issued the New Generation Artificial Intelligence Development Plan in July 2017, setting a three-step goal for the country to become a major global center of AI innovation by 2030 and calling for supporting legal frameworks and ethical guidelines to be developed alongside the technology. China then moved to binding rules ahead of most jurisdictions: its Interim Measures for the Management of Generative AI Services took effect on 15 August 2023, the first national law aimed specifically at generative AI. Governance is not only about software, either. The United States used export controls, beginning with the October 2022 advanced-computing rule, to restrict China’s access to the chips that train large models, making compute itself a lever of state policy.
Governments have also coordinated through a multilateral track that runs from non-binding to binding. The summit series that began at Bletchley Park continued at the Seoul summit in 2024, where sixteen frontier labs signed voluntary safety commitments, and the Paris summit in 2025, whose statement the United States and United Kingdom declined to sign. The G7’s Hiroshima Process produced a voluntary international code of conduct for advanced-AI developers in October 2023, and the United Kingdom stood up the first state-backed AI safety institute the following month. The track reached binding law with the Council of Europe’s Framework Convention on Artificial Intelligence, opened for signature on 5 September 2024, the first legally binding international treaty on AI.
Finally, governments have coordinated internationally through non-binding declarations. At the AI Safety Summit held at Bletchley Park on 1 to 2 November 2023, the United Kingdom convened countries that signed the Bletchley Declaration, which states that AI “should be designed, developed, deployed, and used, in a manner that is safe” and that the most capable frontier systems carry a “potential for serious, even catastrophic, harm.” Such declarations express shared intent and commit signatories to cooperation, including the establishment of AI safety institutes and research networks, but do not by themselves impose legal obligations.