The EU AI Act is the European Union’s law regulating artificial intelligence, the first comprehensive AI law of any major jurisdiction. Adopted in 2024, it sorts AI systems by the risks their uses pose, bans a small set of practices outright, imposes strict requirements on “high-risk” applications and sets special rules for the most powerful general-purpose models. Its obligations take effect in stages, mostly by August 2028.
Regulation by risk
The Act’s core idea is to regulate mostly uses rather than technologies; the rules for general-purpose models are the main exception:
- Unacceptable risk: Banned since February 2025, including social scoring by public or private actors, manipulative techniques that exploit vulnerabilities, building facial-recognition databases by untargeted scraping of images, emotion recognition at work and in schools, and most real-time remote biometric identification in public spaces by police.
- High risk: AI used in areas such as hiring, credit scoring, education, critical infrastructure, law enforcement, migration and the administration of justice, or as a safety component of regulated products. Providers must manage risks, use quality data, keep logs, ensure human oversight and register their systems.
- Transparency duties: People must be told when they interact with a chatbot, and, from August 2026 (December 2026 for systems already on the market), AI-generated content must be marked in a machine-readable way (see AI labeling and watermarking).
- Minimal risk: Most AI applications, such as spam filters or video games, face no new obligations.
The rules for general-purpose AI models apply since August 2025. All providers must adopt a copyright policy that respects opt-outs and publish a summary of their training data, and most must also document their models (open-source models get lighter duties); providers of models with “systemic risk” – presumed above 10^25 floating-point operations of training compute – must also evaluate and mitigate risks, report serious incidents and ensure cybersecurity. A voluntary code of practice published in July 2025 explains how to comply; most large US and European developers signed it, Meta declined, and xAI signed only the safety chapter.
Revision before full force
Even before most obligations took effect, the EU began to revise the Act. Responding to complaints from industry and member states about complexity and missing technical standards, and to a report by former European Central Bank president Mario Draghi on Europe’s lagging competitiveness, the Commission proposed a “digital omnibus” in November 2025. The amendments, in force since July 2026, postpone the high-risk obligations to December 2027 for most systems and August 2028 for those built into regulated products, soften the duty to promote AI literacy and add a ban, applying from December 2, 2026, on AI systems that generate non-consensual intimate images (“nudifiers”) or child sexual abuse material (see nudification). Civil-society groups such as EDRi and ECNL called it deregulation at the expense of fundamental rights, and some linked it to pressure from Washington and Big Tech; some industry groups said it did not go far enough.
Assessments
Supporters see the Act as a model of rights-based, democratic AI governance that could set global standards through the Brussels effect. Critics, including many in Europe’s tech industry, argue that it is complex and burdensome, that it regulates before risks are understood, and that it may drive investment and model releases away from Europe; some companies have delayed launching AI features in the EU, though mostly citing the GDPR or the Digital Markets Act rather than the AI Act itself. From a classical-liberal standpoint, its general focus on concrete uses rather than research is a strength, while its breadth and bureaucracy are its weakness.
Why it matters
The AI Act is the most ambitious attempt so far to put AI under democratic law. How it works in practice – whether it protects people without smothering innovation – will shape not only European AI but the global debate about how far governments should go, and its rapid revision shows how hard it is to legislate for a fast-moving technology.
Written with Claude, made by Anthropic, which signed the EU’s code of practice for general-purpose AI and is subject to the Act’s rules for such models.
Related
Brussels Effect · AI Labeling · Regulatory Sandbox · Explainability · Precautionary Principle · Europe Just Watched the Off-Switch Work
Sources: Regulation (EU) 2024/1689 (EUR-Lex) (June 13, 2024) · EU AI Act, Article 5 · EU AI Act, Article 51 · European Commission: The General-Purpose AI Code of Practice (July 10, 2025) · Mario Draghi: The future of European competitiveness (European Commission) (September 9, 2024) · Lewis Silkin: The Digital Omnibus on AI enters into force today (July 27, 2026) · Gibson Dunn: EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes (May 27, 2026) · Al Jazeera: EU moves to ease AI, privacy rules amid pressure from Big Tech, Trump (November 20, 2025) · ECNL: AI Omnibus – deregulation at the expense of fundamental rights (May 11, 2026)
Leave a Reply