EU AI Act Undergoes Significant Changes
EU AI Act Undergoes Significant Changes
https://www.wsgr.com/en/insights/eu-ai-act-undergoes-significant-changes.html
Publish Date: 2026-05-08 13:46:00
Source Domain: www.wsgr.com
-
Delay in High-Risk AI System Obligations: The requirements for high-risk AI (HRAI) systems originally set to apply from August 2026 have been rescheduled to December 2027. HRAI systems covered by EU product safety legislation will have until August 2028.
-
Industrial AI Carve-Out: Industrial AI embedded in machinery will now follow EU machinery rules instead of the AI Act’s HRAI regime, providing more flexibility for these systems.
-
Transparency Obligations Postponed: The obligation to mark AI-generated content will now apply from December 2026, extending the initial effective date by four months.
-
New Prohibitions on Harmful Content: The amended AI Act will include two more categories of prohibited AI systems: those generating non-consensual sexualized deepfakes and those creating child sexual abuse material.
-
Maintained Registration Obligation: The obligation to register high-risk AI systems remains in place, regardless of exemptions.
-
Narrowing the Definition of Safety Components: For safety-critical applications, the definition of “safety component” has been narrowed to exclude AI systems whose failure or malfunction does not pose health or safety risks.
-
Broader Oversight Powers: The EU Commission’s AI Office will extend its oversight to AI systems built on general-purpose AI models, those in large online platforms, and large search engines.
-
Use of Sensitive Data: The deal allows sensitive data, such as race and ethnicity, to be used strictly as necessary for bias detection and correction across all AI systems, not just high-risk ones.
