Global AI Law Index

Analysis: This bill, which would give the Federal Trade Commission (FTC) sweeping powers at the federal level, requires companies to conduct an "Impact Assessment" of the decision-making algorithms they use. Its aim is transparent reporting on whether those algorithms discriminate by race, gender, or age, particularly in areas that directly affect people's lives, such as recruitment, lending, housing, and education.

Set against the general US reluctance to regulate, the bill is the most concrete legislative step yet toward addressing how "black box" systems harm consumers. If it becomes law, tech companies will have to explain to the FTC how their algorithms are trained.

Analysis: The world's first comprehensive, legally binding horizontal artificial intelligence law. It was adopted in June 2024 and entered into force in August 2024, with a phased transition. The law sorts AI into four tiers according to the risk it creates (the risk-based approach) rather than the technology itself: Unacceptable Risk, High Risk, Limited Risk, and Minimal Risk.

The law's most significant feature is the "Brussels Effect": a US or Chinese company that wants to sell a product in the European market must comply with its rules. Social scoring, subliminal manipulation, and unauthorized biometric categorization are banned outright, while General Purpose AI (GPAI) and Large Language Models (LLMs) are subject to transparency, copyright compliance, and systemic risk reporting obligations.

Analysis: AIDA, part of Canada's Bill C-27, departs from the European model and follows the logic of "Agile Regulation." Rather than regulating AI systems in detail, it sets a framework only for "High-Impact" systems and leaves the details to future regulations.

AIDA's most striking feature is a new executive office, the "AI and Data Commissioner." Although the bill initially favored guidance so as not to stifle innovation, it stood out for proposing to bring malicious and harmful uses of AI directly within the criminal law, with heavy penalties including imprisonment.

Analysis: This bill, the Global South's boldest step in technology regulation, combines the risk-based approach of the EU AI Act with Brazil's own case law. It contains particularly strong provisions on the labor market, racial discrimination, and the right to challenge algorithmic decisions (Human in the Loop).

The most notable aspect of the Brazilian bill is its clear position on copyright. It proposes new mechanisms under which AI companies would pay royalties to content creators for copyrighted works used in data mining and AI training. As introduced in 2023, it also made providers and operators of high-risk and excessive-risk AI systems strictly liable for the damage they cause.

Analysis: Agreed by 28 countries and the European Union, including the US, China and Turkey, at the summit held in November 2023 at Bletchley Park (UK), where the Enigma codes were broken during WWII; New Zealand joined in October 2024. Although it is not legally binding, it marked a turning point in global AI diplomacy.

The declaration focuses on the risks posed by "Frontier AI" models, particularly in domains such as cybersecurity and biotechnology, while also acknowledging everyday harms such as bias, privacy and data protection. Because geopolitical rivals such as China and the US signed the same text, it became the first global agreement to record officially that AI safety is a transnational challenge that transcends politics.

Analysis: Published by the White House Office of Science and Technology Policy (OSTP), this document does not have the force of law, but it sets out the philosophical foundation of the US government's approach to AI. It is built around five core principles: Safe and Effective Systems, Algorithmic Discrimination Protections, Data Privacy, Notice and Explanation, and Human Alternatives, Consideration, and Fallback.

Its most significant element is the emphasis on the "Opt-Out" right and on the citizen's right to speak to a human rather than a bot (Human Fallback). A non-binding OSTP framework published in October 2022, it influenced agency guidance under the previous administration; it was archived after the change of administration in January 2025 and no longer reflects federal policy.

Analysis: It is the "world's first legally binding international treaty" on artificial intelligence. Whereas the EU AI Act is essentially a market regulation, the Council of Europe convention is written from the perspective of human rights, democracy, and the rule of law.

Opened for signature in September 2024, the convention obliges signatory states to enact national laws that prevent AI from undermining the judicial system, democratic processes (e.g., through election manipulation), and human rights. The main point of contention is that national security and defense applications are excluded from the convention's scope altogether.

Analysis: Signed by US President Joe Biden in October 2023 and revoked on 20 January 2025 by Executive Order 14148, this Executive Order (EO) directed federal agencies across government (Defense, Energy, Commerce, Health) to set AI standards, as a way around a slow-moving Congress. It is the clearest sign that the US has chosen a sectoral, distributed approach to AI governance over a single central law.

The most striking legal mechanism of the EO was its use of the Defense Production Act to require companies developing powerful AI models (such as OpenAI and Google) to report to the US government, among other things, the results of their safety testing (Red-Teaming). Since its revocation, US federal AI policy has been shaped by EO 14179 (January 2025), America's AI Action Plan (July 2025) and EO 14365 (December 2025) on a national policy framework for AI.

Analysis: In force since August 2023, this regulation is one of the world's first legal texts to regulate "Generative AI" specifically. Enforced by the Cyberspace Administration of China (CAC), it is built on "Content Control and Ideological Integrity," a sharp departure from the West's risk-based approach.

Article 4 stipulates that generative AI must "adhere to core Socialist values" and not generate content aimed at subverting state power or disrupting national unity. It also requires AI providers to file their algorithms in a state database and to undergo security assessments. While it allows flexibility for research, it sets up a strict state censorship and surveillance infrastructure for public-facing uses.

Analysis: A narrowly targeted law (2021/2022) that regulates the "Recommendation Algorithms" underlying platforms such as TikTok (Douyin) and Taobao, not just AI models. It is a direct state intervention against algorithmic designs that exploit users' attention and foster addiction.

The law requires companies to give users the right to "completely turn off the algorithmic recommendation system." It also expressly prohibits programming algorithms in ways that exploit platform workers (couriers, drivers) or endanger their physical health. Using algorithms for excessive pricing or anti-competitive (monopolistic) conduct attracts administrative sanctions.

Analysis: Adopted in May 2019 and updated in 2024, these principles were the first intergovernmental standard on artificial intelligence. They have been adopted by over 40 countries, including the US, the EU, and Turkey. Although they are not legally binding (Hard Law), they serve as the primary blueprint (Soft Law) for national AI strategies worldwide.

They rest on five core principles: Inclusive growth/sustainable development, Human-centered values/fairness, Transparency/Explainability, Robustness/Security, and Accountability. The fact that even the AI definition in the EU AI Act is taken directly from the technical definition formulated by the OECD (2023 update) shows how influential these principles are in international law.

Analysis: The main strategy document showing that, after Brexit, the UK has parted ways with the EU and drawn up its own digital roadmap. Rejecting the EU's rigid, centralized "AI Act" model, the UK adopts a "Context-specific" and "Pro-innovation" approach rather than creating a new AI regulator. The government confirmed this approach in its February 2024 response to the White Paper consultation.

Under this approach, the existing health regulator (MHRA) will regulate AI used in healthcare, and the financial regulator (FCA) AI used in finance, each on the basis of its own principles. Instead of imposing heavy bans, the government sets 5 core principles (Safety, Transparency, Fairness, Accountability, Contestability) and builds a "light-touch" legal framework that encourages companies to innovate in regulatory sandboxes.

Analysis: Adopted by UNESCO's 193 member states in 2021, it is the global standard on AI ethics with the widest participation. It is a normative instrument that addresses not only technological development but also AI's effects on cultural diversity, the environment, education, and gender equality.

The recommendation advises states to establish "Ethical Impact Assessment" methodologies. At its core lies the philosophy that "Technological development cannot proceed at the expense of human rights and nature." Although it is not legally binding, it provides a direct blueprint for draft legislation in countries, for example in Africa and Latin America, that have not yet enacted their own AI laws.