The EU Rewrites Its AI Act: A Relaxation Under Pressure

The EU Revises Its AI Act Before Full Implementation

Less than two years after its adoption, the European regulation on artificial intelligence, known as the AI Act, has been revised. On May 7, 2026, the European Parliament and the EU Council reached an agreement on the "Digital Omnibus on AI," a text proposed by the European Commission five months earlier. This text aims to simplify the rules for businesses, a balance praised by Vice President Henna Virkkunen between innovation and citizen protection.

New Deadlines and Easing for Businesses

Two new deadlines have been set. High-risk autonomous AI systems must comply by December 2, 2027. These systems cover areas such as biometrics, education, employment, migration, and border control. For systems integrated into regulated products like elevators, toys, and machinery, the deadline is extended to August 2, 2028.

Medium-sized enterprises also benefit from this easing. Privileges previously reserved for SMEs, such as simplified documentation and scaled sanctions, are now extended to companies employing fewer than 750 employees. A European regulatory sandbox will allow for testing systems in real-world conditions.

Bans and Expanded Powers

The agreement adds a new prohibition to Article 5 of the AI Act, banning AI "nudification" systems and artificially generated child sexual abuse content. The European AI Office sees its oversight powers expanded to include general-purpose models.

Challenging Negotiations

Discussions around these revisions have not been straightforward. The trilogue on April 28 ended in a deadlock after twelve hours of negotiations, primarily due to the complex interplay between the AI Act and existing sectoral regulations. This point of friction has fueled criticism since the summer of 2025.

A Rapid and Unprecedented Revision

Unlike the GDPR, which took nine years to be amended, the AI Act was reopened just sixteen months after its publication in the Official Journal. This speed is unprecedented for a structural EU regulation.

Industrial pressure played a major role. In July 2025, 45 major companies, including Airbus and Axa, called for a pause, followed by 56 AI stakeholders, led by Mistral AI, demanding simplification. The Draghi report estimated the cost of regulatory compliance in Europe at €500 billion per year.

While Brussels eases its legislation, Washington still lacks a federal framework on AI. In January 2025, the Trump administration signed an executive order to "remove barriers" to technological development, although recent statements indicate a backtrack in light of the risks posed by frontier models like Claude Mythos.

In France, Mistral AI, valued at €11.7 billion with 350 employees, anticipated these changes by publishing a memo of 22 proposals in April. For French startups, this easing provides a concrete 16-month reprieve to comply with high-risk rules. However, the investment gap between Europe and the United States remains significant, with $8 billion invested on the European side compared to $109 billion in the United States.