EU AI Act Amendments: New Proposals from the European Commission

24.11.2025

On 19 November 2025, as part of the Digital Omnibus package, the European Commission presented a draft set of amendments to the EU AI Act. 

As a reminder, the EU AI Act (the “Act”) is the world’s first comprehensive EU legislative framework regulating the development and use of artificial intelligence systems.

Purpose of the amendments: to facilitate a timely, smooth, and proportionate implementation of certain provisions of the Act, thereby improving the functioning of the internal market.

The draft amendments have not yet been adopted – they will undergo readings in the European Parliament and negotiations with EU Member States.

What key changes have been proposed?

• Article 4 – AI literacy: instead of placing obligations on AI providers and developers, the responsibility for ensuring sufficient AI literacy is shifted to the European Commission and EU Member States.
• New Article 4a (replacing Article 10(5)) – Processing special categories of personal data to detect and mitigate bias: providers and developers of AI systems and models may be permitted to process special categories of personal data for the purpose of identifying and correcting bias, subject to specific safeguards.
• Article 6 – Classification rules for high-risk AI systems: previously, a provider who believed that an AI system listed in Annex III (e.g., remote biometric identification systems, emotion recognition systems, etc.) was not high-risk had to document its assessment and register the system. Under the proposed changes, registration in the EU database would no longer be required.
• Article 11 – Technical documentation: companies classified as SMCs (up to 750 employees and turnover not exceeding EUR 150 million) and SMEs (up to 250 employees and turnover not exceeding EUR 50 million) may submit technical documentation (Annex IV) in a simplified form. The European Commission will be responsible for developing this simplified format.
• Article 17 – Quality management system: the requirements for a quality management system should take into account the capabilities and constraints of SMCs and SMEs. Some elements may be implemented in a simplified manner.
• Article 57 – AI regulatory sandboxes: the AI Office will be empowered to establish EU-level AI regulatory sandboxes for certain AI systems, offering priority access to SMCs and SMEs, and to require EU Member States to strengthen cross-border cooperation between their national sandboxes.
• Article 75 – Mutual assistance, market surveillance, and oversight of general-purpose AI systems: the powers of the AI Office would be expanded. It would gain authority to oversee AI systems based on general-purpose AI models when both the model and the final product are developed by the same provider.
• Article 99 – Fines: reduced administrative fines would be introduced for SMCs. Currently, reduced fines apply only to SMEs.
• New Annex XIV: will contain a list of codes, categories, and corresponding AI system types for the purpose of the notification procedure (Article 30).

Article 113 used to set the general date of application of the Act to 2 August 2026. However, delays in preparing standards, specifications, and establishing national authorities make it difficult to meet the high-risk AI compliance requirements by this date.

This means that the compliance timelines for companies developing and using AI systems will change. Once the European Commission confirms the availability of the necessary support measures, the rules will apply:

• after 6 months – for high-risk systems where AI is used in sensitive or critical application areas;
• after 12 months – for high-risk systems where AI functions as a safety component of a product.

However, the maximum application deadlines are set at 2 December 2027 and 2 August 2028, respectively – by these dates the rules will apply in any case.

You can access the full text of the draft amendments to the EU AI Act at the provided link.

---

Attention Journalists: Use of REVERA website materials in publications is only allowed with our written permission.