The EU Slows Down AI Regulation

Why the world's first AI law ended up slowing itself down

March 6, 2026 | Kim Kyung-jin

First, the one-line summary

The EU created the world's first AI law in 2024. But at the end of 2025, it changed direction and said it would "move a little more slowly." The reason was that the law had gone too far ahead, and the system was not ready to comply.

Why did it change direction? Two reasons

First, money. Comparing AI investment in 2024, the United States invested $109 billion, while the entire EU invested less than $20 billion. That is a gap of more than five times. A sense of crisis grew that the stricter the regulations became, the more companies would invest in the United States rather than the EU.

Second, lack of preparation. The law had been made, but the things needed to comply with it were not yet in place. EU standardization bodies failed to create technical standards by 2025. Most countries had not designated supervisory authorities. The Commission also failed to release guidance on time. In short, the rules existed, but no one yet knew exactly what those rules required.

Digital Omnibus: The EU's Proposed Solution

On November 19, 2025, the European Commission announced a legislative amendment package named 'Digital Omnibus'. Its goal is to reduce corporate burden by at least 25%, and by at least 35% for small and medium-sized enterprises. It revised not only the AI Act, but also the General Data Protection Regulation (GDPR) and cybersecurity-related laws.

What exactly changes? Five points

•1. Enforcement of high-risk AI rules delayed by up to 16 months

High-risk AI that directly affects people, such as hiring screening, emotion recognition, and credit scoring, was originally set to face strict regulation from August 2026. That will be pushed back to December 2027. AI directly tied to product safety, such as AI embedded in medical devices or aviation systems, will be extended until August 2028.

•2. Six-month grace period for labeling AI-generated content

Application of the rule requiring watermarks or tags on AI-generated text, images, and videos will be delayed, for services released before August 2026, until February 2027.

•3. Responsibility for AI literacy training shifts

Previously, companies that made or used AI services were responsible for employee training. Going forward, the Commission and national governments will take on that responsibility. The burden on companies will fall.

•4. AI oversight authority concentrated in the AI Office

At present, AI oversight authority is divided between the EU's AI Office and national agencies. Going forward, all oversight related to general-purpose AI models will be handled by the AI Office alone. From a company's point of view, there will be no need to deal separately with agencies in multiple countries.

•5. Requirements eased for using personal data in AI development

When AI uses sensitive personal data to detect and correct bias, it was previously allowed only when 'strictly necessary.' This will be eased to 'necessary.' A new legal basis for using personal data as AI training data may also be added.

What does not change

This easing is a timing adjustment, not a change in the direction of the law. The basic system of classifying AI by risk level, transparency obligations for chatbots and deepfakes, obligations for general-purpose AI models (already in force as of August 2025), and the penalty system remain intact.

Where things stand now: Passage is still undecided

The Digital Omnibus is not yet final law. It must be approved by the European Parliament and the Council of the EU before final passage. In February 2026, the Parliament's co-rapporteurs released a draft amendment, and discussions are under way. If it is not passed before August 2, 2026, enforcement will begin under the original timeline. Companies are safer preparing now rather than waiting for the law to pass.

The lesson for Korea

The EU's experience teaches one thing. Making a law and making it possible to comply with that law are different matters. Korea's AI Framework Act also took effect on January 22, 2026, and, like the EU, it includes a guidance period. To avoid following the EU's path, technical standards and supervisory authorities must be ready alongside enforcement of the law. That is why the role of AI law and compliance experts matters so much right now.