Artificial Intelligence: Council and Parliament agree to simplify and streamline regulation – Consilium and Proposition de règlement du Parlement européen et du Conseil, 13 mai 2026
Less than two years after the adoption of Regulation 2024/1689 ‘laying down harmonised rules on artificial intelligence’ (‘AI Regulation’), and whilst it is not yet fully applicable, the European Union is already embarking on the simplification of this text.
On May 7th, 2026, the Council of the European Union and the European Parliament reached a provisional political agreement aimed at simplifying certain obligations set out in the European Regulation on Artificial Intelligence. This reform forms part of the ‘Digital Omnibus’, a broader initiative designed to ease certain administrative burdens on European economic operators.
A clear commitment to regulatory simplification
The AI Regulation is the world’s first comprehensive regulatory framework for artificial intelligence. However, its operational complexity and the Commission’s delay in publishing certain guidelines quickly raised concerns, particularly among European businesses facing particularly onerous obligations regarding documentation, governance, compliance, risk assessment and human oversight.
The political agreement of May 7th,2026 thus seeks to give businesses some breathing space in order to prevent excessive regulatory complexity from stifling European innovation or creating excessive legal uncertainty for industry players.
Deferral of certain obligations applicable to high-risk systems
The main change concerns the timetable for the application of obligations applicable to high-risk AI systems.
According to the provisional agreement:
- the obligations relating to high-risk systems covered by Annex III of the AI Regulation, which were due to come into force on August 2nd, 2026, would be postponed to December 2nd, 2027;
- those applicable to systems incorporated into regulated products (Annex I of the AI Regulation), scheduled for August 2nd,2027, would be postponed to August 2nd, 2028.
The grace period granted to suppliers who have already placed their systems on the market before August 2nd, 2026 to comply with the marking requirements set out in Article 50 of the RIA would now be set at four months.
Finally, the deadline for the competent national authorities to establish regulatory AI sandboxes has been extended by one year, to August 2nd, 2027.
In practice, this postponement is intended to give businesses and national authorities more time to put in place the compliance mechanisms required by the regulation.
Relief from obligations for certain sectors
The compromise also provides a fundamental clarification of the relationship between the AI Regulation and sector-specific regulations. It provides for the possibility of limiting the application of certain obligations under the Regulation where sector-specific legislation listed in Section A of Annex I to the AI Regulation already ensures an equivalent level of protection in terms of health, safety or fundamental rights.
This approach aims to avoid regulatory overlaps and reduce compliance burdens, whilst giving the Commission a central role in identifying such situations, notably through the adoption of delegated acts. It is accompanied by specific treatment for certain industrial sectors, in particular the machinery sector, to ensure greater consistency in the applicable regulatory framework.
The ban on new AI practices
The ongoing discussions are not merely aimed at relaxing the text: they also provide an opportunity to expand the list of prohibited practices in Article 5 of the AI Regulation in light of the rapid evolution of generative uses that most infringe upon fundamental rights.
The co-legislators plan to add an explicit ban on AI systems designed to generate or manipulate non-consensual sexual or intimate content, as well as systems producing content relating to child sexual abuse. The recitals of the text specifically target the dissemination of ‘nudification’ applications and justify this intervention on the grounds of the serious risks that such uses pose to the dignity, privacy and integrity of individuals and, in the case of minors, to children’s rights.
Extension of proportionality measures to medium-sized companies
The compromise extends certain proportionality measures, which were initially intended to reduce the administrative burden on SMCs, to small and medium-sized enterprises (SMEs). This is aimed in particular at facilitating the transition of SMEs towards the obligations applicable to large enterprises.
Continuation of the phased transition
This political agreement was swiftly translated into legal terms through the formalisation of the compromise in a proposal for a regulation of the European Parliament and of the Council of May 13th, 2026, which sets out the procedures for its integration into the existing framework.
The provisional agreement will then need to be approved by both European institutions with a view to the formal adoption of the legislative act in the coming weeks.
The other obligations set out in the AI Act remain unchanged, in particular those relating to the transparency of AI systems. Compliance work must therefore, in any event, continue. In this regard, the European Commission published, the very day after the political agreement was reached, draft guidelines on the transparency obligations set out in Article 50 of the AI Act. These obligations require, in particular, that individuals be informed when they interact with an AI system, when they are exposed to artificially generated content, or when a system uses emotion recognition or biometric categorisation mechanisms.
