2025
EU AI Act takes effect
The EU AI Act enters into force in phases: the first binding international regulation for AI, with prohibitions on high-risk applications and transparency requirements for LLMs.
The world's first AI law
The EU AI Act entered into force on August 1, 2024, with a phased implementation schedule running to 2027. By 2025, the first binding provisions were in effect: prohibitions on unacceptable-risk AI systems (such as real-time mass biometric surveillance in public spaces, social scoring systems, and subliminal manipulation), and requirements for high-risk AI systems in areas such as critical infrastructure, education, employment, and law enforcement.
What the AI Act regulates
The AI Act uses a risk-based approach with four categories. Unacceptable risk: banned outright. High risk: allowed but subject to strict requirements including conformity assessments, transparency, human oversight, and registration in an EU database. Limited risk: transparency obligations (chatbots must disclose they are AI). Minimal risk: no specific requirements. For general-purpose AI models (GPAIMs) with more than 10^25 FLOPs of training compute — the frontier model threshold — additional requirements apply, including transparency about training data and adversarial testing.
Global impact
The EU AI Act is the first comprehensive AI legislation anywhere in the world and has significant extraterritorial effect: any AI system used or sold in the EU must comply, regardless of where it was developed. This gives it a Brussels effect similar to GDPR: companies building AI for global markets effectively need to comply with EU standards. The US took a different approach — voluntary commitments and executive orders rather than binding legislation — while China had its own separate regulations for generative AI and recommendation algorithms.
Debate
The AI Act was controversial from the start. AI companies and some researchers argued it would stifle innovation and push AI development outside Europe. Others argued it did not go far enough on the most powerful models. The debate over how to regulate AI without hampering its benefits while preventing its harms is far from resolved — the EU AI Act is the first attempt at a systematic answer, not the last word.