European Commission Defends Policy Decisions in Proposed EU AI Law Changes
On January 14, 2026, the European Commission defended fundamental policy decisions in the proposed amendments to the EU AI Act. These amendments maintain prohibitions on real-time biometric identification in public spaces and enforce risk-based obligations for high-risk AI systems. Commission Executive Vice-President Margrethe Vestager confirmed transparency requirements for general-purpose AI models and established systemic risk thresholds for models training exceeding 10×1025 FLOPs.
Key Amendments and Provisions
The proposed changes aim to address implementation challenges faced by 4,500 SMEs while preserving fundamental rights protections across 27 member states.
- Risk-based framework maintains Article 5 prohibited practices.
- General-purpose AI transparency obligations preserved intact.
- High-risk system conformity assessments streamlined.
- The AI Office gains exclusive supervisory powers.
- SME simplification measures extend technical documentation relief.
- Implementation timeline preserves phased application dates.
- Transparency obligations mandate AI-generated content labeling.
- Codes of practice standardize systemic risk evaluations.
- Market surveillance framework coordinates the efforts of 27 member states.
- Bilateral cooperation facilitates compliance for third countries.
- Impact assessment validates €6.2 billion in compliance savings.
- Legislative procedure targets ordinary adoption in Q3 2026.
Preservation of Article 5 Prohibitions
The amendments ensure that Article 5 prohibitions remain intact, which include:
- Real-time remote biometric identification in public spaces.
- Manipulative subliminal techniques.
- Social scoring by public authorities.
Commission documentation confirmed that 2,400 deployed systems across 18 member states exhibited a 78% non-compliance rate prior to the enforcement commencement in August 2024.
Streamlined Conformity Assessments
High-risk AI systems classification continues under Annex III, which now includes:
- Biometric categorization.
- Critical infrastructure.
- Employment and education systems.
- Biometric safety components.
Commission impact assessments project a 24-month transition period for legacy systems and aim to designate 85 conformity assessment bodies by Q3 2026.
Transparency Obligations for General-Purpose AI
Chapter V obligations for general-purpose AI models remain unchanged, requiring detailed summaries of:
- Training data sources.
- Quality evaluation results.
- Systemic risk assessments.
Models exceeding 10×1025 FLOPs must also report any cybersecurity incidents to the AI Office within 72 hours.
Exclusive Supervisory Competence for the AI Office
The amendments grant the European AI Office exclusive competence over high-risk systems that integrate general-purpose AI models developed by the same provider. This centralizes oversight of AI systems embedded in Very Large Online Platforms as designated under the Digital Services Act.
Support for Small and Medium Enterprises
Small and medium enterprises will benefit from simplified technical documentation requirements, reducing compliance costs by 42% for 3,800 high-risk deployments. Additionally, regulatory sandboxes support 1,650 SMEs testing high-risk systems in controlled environments.
Implementation Timeline
The amendments preserve the established legislative calendar, with:
- High-risk AI systems under Annex III applying from December 2, 2027.
- General-purpose AI obligations effective from August 2, 2025.
Conclusion
The proposed amendments aim to balance innovation with fundamental rights protection across the EU market. With a focus on transparency and accountability, the EU AI Act is set to pave the way for a human-centric and trustworthy AI landscape.
