The European Union’s AI Act, which entered into force in August 2024, is the world’s first comprehensive legal framework for artificial intelligence. It creates a risk-based classification system: unacceptable risk (prohibited outright, including social scoring and real-time biometric surveillance in public spaces), high risk (subject to strict requirements including risk assessments, human oversight and transparency), limited risk (transparency obligations only) and minimal risk (no additional requirements). Most generative AI systems, including large language models, fall into a new category of “general-purpose AI” with specific obligations around training data disclosure, copyright compliance and systemic risk assessment.
The US approach is less centralised. President Biden’s October 2023 Executive Order on AI required developers of the most powerful AI systems to share safety test results with the government, directed NIST to develop AI safety standards and called on federal agencies to assess AI’s impact on their domains. But executive orders are not legislation — they can be rescinded by a subsequent president. Congress has held dozens of hearings on AI regulation but has passed no major AI-specific legislation, leaving the US with a patchwork of agency-level guidance rather than a unified legal framework.
China’s approach is different again. Its 2023 “Interim Measures for the Management of Generative AI Services” requires AI systems to “adhere to the core values of socialism,” not generate content that threatens national security, and label AI-generated content. Chinese regulators have also imposed algorithmic recommendation rules that require platforms to give users control over their recommendation algorithms — a surprising consumer protection move driven as much by concerns about social stability as by consumer rights. China’s AI regulation is enforceable and has real penalties, but it is primarily oriented around state control rather than individual rights.
The global fragmentation of AI regulation creates a compliance headache for companies operating internationally. A model that is legal in the US might require modification for the EU and might not be permitted at all in China. The OECD’s AI Principles, adopted by 47 countries, provide a voluntary framework for alignment, but they lack enforcement mechanisms. The next few years will determine whether global AI governance converges toward common standards or fractures along geopolitical lines.