Panorama réglementaire mondial

La réglementation de l'IA dans le monde

Comment le règlement IA européen se compare aux cadres réglementaires d'autres juridictions.

Verified against Regulation (EU) 2024/1689 — 27 August 2026

Le moteur d'AZComply couvre intégralement le règlement IA — la réglementation de l'IA la plus complète au monde.

European Union

EU AI Act

En vigueur

Regulation (EU) 2024/1689 — the world's first comprehensive, binding AI legal framework. Establishes a risk-based classification system covering all AI systems placed on the EU market or affecting EU persons. As amended by the Digital Omnibus (Regulation (EU) 2026/1744, in force 27 July 2026), which extended high-risk deadlines and added new prohibited practices.

Date d'entrée en vigueur: August 1, 2024

Approche par les risques

Four-tier risk pyramid: Prohibited (Article 5), High-Risk (Annex III), Limited Risk (Article 50), Minimal Risk. Plus a separate GPAI track for foundation models.

Sanctions

Up to €35M or 7% of global annual turnover for prohibited practices; up to €15M or 3% for high-risk violations; up to €7.5M or 1.5% for misinformation to authorities.

Portée extraterritoriale

Strong extraterritorial reach — applies when AI output is used in the EU, regardless of provider location.

Caractéristiques clés

Deterministic risk-based classification across 4 levels
Amended by Digital Omnibus (Regulation (EU) 2026/1744, in force 27 July 2026) — Annex III high-risk compliance 2 Dec 2027; Annex I product-embedded 2 Aug 2028; new Art. 5.1.i/j prohibitions (nudifier AI, AI-generated CSAM) effective 2 Dec 2026
Annex III lists 8 high-risk sectors including biometrics, employment, critical infrastructure
GPAI obligations with systemic risk threshold at 10^25 FLOPs
Mandatory technical documentation, human oversight, transparency
Fundamental Rights Impact Assessment (Article 27) for deployers
EU AI Office as central supervisory authority
EU-wide public database for high-risk AI systems (Article 71)
AI literacy obligations for all operators (Article 4)

Run Free Check

United Kingdom

UK AI Safety

Volontaire

Pro-innovation, principles-based approach administered by sector-specific regulators. The AI Safety Institute (now DSIT) focuses on frontier model evaluation. No single binding AI Act equivalent; sector regulators (FCA, ICO, CMA) apply existing legislation to AI.

Approche par les risques

No formal tiered risk classification system. Sector regulators assess risk within their own frameworks.

Sanctions

No AI-specific penalties. Enforcement via existing sectoral legislation (FCA, ICO, CMA fines apply in respective domains).

Portée extraterritoriale

Limited extraterritorial reach based on existing UK GDPR and sector-specific rules.

Caractéristiques clés

Pro-innovation stance — no binding AI Act
AI Safety Institute leads frontier model evaluation (Bletchley Park legacy)
Cross-sector AI principles: safety, security, fairness, accountability, contestability, redress
ICO guidance on data protection implications of AI
FCA guidance on AI in financial services
Regulatory sandbox for AI innovation
Voluntary AI Code of Practice for general-purpose AI

United States

US AI Framework

Volontaire

Executive Order 14110 (October 2023) directed federal agencies on safe AI development. NIST AI Risk Management Framework (AI RMF 1.0) provides voluntary guidance. No federal AI Act; sector-specific rules apply (FDA for medical AI, banking regulators for fintech, etc.).

Date d'entrée en vigueur: October 30, 2023

Approche par les risques

No binding federal risk tiers. NIST AI RMF provides voluntary risk management guidance across four functions: Govern, Map, Measure, Manage.

Sanctions

No AI-specific federal penalties. FTC enforcement powers apply to unfair/deceptive AI practices. Sector regulators (FDA, OCC, CFPB) enforce in respective domains.

Portée extraterritoriale

No extraterritorial AI-specific reach. Standard US jurisdictional rules apply.

Caractéristiques clés

EO 14110 directed NIST, agencies on AI safety standards
NIST AI RMF 1.0 (2023) — voluntary governance framework
NIST Generative AI Profile (NIST AI 600-1) for GenAI risks
Sector-specific guidance: FDA for AI/ML medical devices (predetermined change control)
FTC enforcement on deceptive AI claims
State-level legislation active: California SB 1047, Colorado AI Act, Texas HB 149
No federal pre-market approval for most AI systems

China

China AI

En vigueur

China has enacted three major AI regulations: Algorithm Recommendation Management Provisions (2022), Deep Synthesis (Deepfake) Provisions (2023), and Interim Measures for Generative AI Services (2023). Additional draft regulations on facial recognition and AI security assessment are in progress.

Date d'entrée en vigueur: Multiple (2022-2023)

Approche par les risques

Security assessment required for AI services with "public opinion attributes" or "social mobilization capability." Generative AI services face specific content obligations.

Sanctions

Fines up to RMB 100,000 (approx. €13,000) for data-related violations. Potential service suspension and security assessment denial. CAC (Cyberspace Administration of China) enforcement.

Portée extraterritoriale

Applies to services accessed by users in China, including foreign providers.

Caractéristiques clés

Algorithm Recommendation Provisions (2022) — transparency, opt-out rights
Deep Synthesis Provisions (2023) — deepfake labelling, watermarking
Generative AI Interim Measures (2023) — content obligations, security assessment
CAC registration required for public-facing generative AI
Training data must comply with data security and personal information laws
AI must align with "socialist core values"
Security assessment for AI with "public opinion formation" or "social mobilization" capability

Canada

AIDA (C-27)

Proposé

AIDA is Part 3 of Bill C-27 (Digital Charter Implementation Act, 2022). It establishes a risk-based framework for AI systems, focusing on "high-impact AI systems." Bill C-27 stalled in Parliament; the incoming government may revise or replace it.

Approche par les risques

Risk-based: "High-impact" AI systems face core obligations. Definition of "high-impact" to be set by regulation, expected to align broadly with Annex III categories.

Sanctions

Proposed: up to C$25M or 5% of global revenues for intentional violations. C$10M or 3% for non-intentional. Criminal penalties for reckless high-impact AI harm.

Portée extraterritoriale

Applies in the context of international and interprovincial trade. Foreign providers offering services in Canada would be captured.

Caractéristiques clés

Part of Bill C-27 alongside CPPA (consumer privacy reform)
Focus on high-impact AI systems — definition by regulation
Mandatory impact assessment, risk mitigation, and anomaly monitoring
Transparency obligations including making assessments publicly available
Designated AI systems face additional human oversight requirements
AI and Data Commissioner to be established
Criminal penalties for reckless deployment causing serious harm

Brazil

Brazil AI (PL 2338)

Proposé

Bill 2338/2023, approved by the Brazilian Senate in November 2024. Now pending in the Chamber of Deputies. Risk-based approach inspired by the EU AI Act, with adaptations for Brazil's digital economy context and constitutional framework.

Approche par les risques

Two-tier risk system: excessive-risk (prohibited) and high-risk systems face stringent obligations. Risk categories broadly align with EU AI Act sectors.

Sanctions

Proposed: up to R$50M (approx. €9M) or 2% of Brazilian revenues. Stricter penalties for systemic violations. National Data Protection Authority (ANPD) enforcement.

Portée extraterritoriale

Applies when AI output affects persons located in Brazil, regardless of where processing occurs.

Caractéristiques clés

Approved by Senate November 2024; pending Chamber of Deputies
Two risk tiers: excessive-risk (prohibited) and high-risk
Algorithmic Impact Assessment (AIA) for public sector high-risk AI
Transparency: users must be informed of AI interaction
Human review right for automated decisions affecting rights
ANPD (data protection authority) as primary AI enforcer
Special protections for vulnerable groups including children and workers

Regulatory landscape comparison — information only. This page describes publicly available regulatory frameworks and does not constitute legal advice. Regulatory status and requirements change frequently; verify current requirements with qualified counsel.