What happened
EU countries and Parliament agreed May 7 to delay enforcement of AI Act rules for high-risk systems—including biometrics, critical infrastructure, law enforcement, education, and employment—from August 2, 2026 to December 2, 2027. The 16-month delay also excludes machinery from the AI Act scope, subjecting industrial AI only to sectoral rules. Mandatory watermarking of AI-generated content will apply from December 2, 2026, and a ban on non-consensual sexually explicit deepfakes was added.
Why it matters
This is the first material rollback of EU digital rules under pressure from industry and member states (led by Germany). The delay grants enterprises breathing room but increases regulatory divergence risk—only a few jurisdictions worldwide have followed the EU AI Act model. Organizations with multi-jurisdiction AI deployments must now track three timelines: general AI Act provisions (in force August 2024), watermarking/deepfake ban (December 2026), and high-risk system requirements (December 2027).
Action needed
Review AI Act compliance roadmaps and adjust internal timelines for high-risk system classification, risk assessments, and conformity processes. Enterprises operating industrial AI (machinery exemption) should confirm whether sectoral rules provide equivalent coverage or create compliance gaps. Monitor formal endorsement by EU governments and Parliament in coming months.