The definitive guide to worldwide AI governance in 2025
As of December 2025, the global AI landscape has shifted toward a risk-based regulatory model, with 83 jurisdictions establishing formal frameworks to balance innovation and safety. This comprehensive report tracks the transition from voluntary ethics to binding legislation, notably led by the implementation of the EU AI Act and diverse national strategies across five continents. The findings reveal a growing international consensus on transparency, data protection, and the criminalization of AI misuse, such as deepfakes.
Points clés
- The report covers a total of 83 jurisdictions, ranging from major powers like the US and China to emerging digital leaders like Rwanda.
- The European Union’s AI Act (Regulation 2024/1689) serves as a global keystone, introducing a risk-based hierarchy and strict conformity assessments.
- China has implemented specific administrative measures via the Cyberspace Administration of China (CAC) focusing on algorithmic recommendations and content labeling.
- Latin American leadership is notable, with Uruguay becoming the first in the region to sign the Council of Europe AI Convention.
- Criminal law is evolving rapidly, with Chile, Peru, and Latvia enacting specific statutes to penalize AI-enabled identity theft and electoral deepfakes.
- Egypt aims for a 7.7% ICT sector contribution to its GDP by 2030 through its Second Edition National AI Strategy.
- Spain has pioneered enforcement by creating AESIA, the first national AI supervision agency in Europe.
- The United States maintains a distributed model, utilizing White House Executive Order 14110 and state-level laws like Tennessee’s ELVIS Act.
- Costa Rica stands out for transparency, having partially drafted its AI Regulation Bill using GPT-4 to invite public scrutiny.
- Over 15 countries, including Saudi Arabia and Singapore, have released specific guidelines for generative AI and large language models (LLMs).
À retenir
If you thought your company could just “move fast and break things” with AI, welcome to 2025—where “breaking things” now comes with a side of heavy fines and potential jail time across 83 different flavors of bureaucracy. Navigating this mess requires more than just a lawyer; you probably need a crystal ball and a map of every sandbox from Denmark to Malaysia. My advice? Just label everything as “high risk” and hope the regulators appreciate your honesty, or perhaps just move your headquarters to a deserted island that hasn’t discovered the internet yet.
Sources
Quiz sur le document: 10 questions
