EU pushes back start date for high-risk AI rules in surprise deal

EU pushes back start date for high-risk AI rules in surprise deal
Credit: Unsplash

EU member states and European Parliament negotiators have agreed a provisional deal to simplify parts of the EU’s rules on artificial intelligence, including delaying when key requirements for “high-risk” AI systems will apply.

The agreement was reached between the Council presidency and Parliament as part of the EU’s “Omnibus VII” package, which the Council described as a set of proposals under the bloc’s simplification agenda for digital rules and the implementation of harmonised AI requirements, the Council of the EU informed on Thursday.

Under the provisional deal, the start date for rules on high-risk AI systems would be pushed back to 2 December 2027 for stand-alone high-risk systems and to 2 August 2028 for high-risk AI built into products.

High-risk systems are those subject to tighter obligations under the AI Act because of their potential impact in areas such as safety and fundamental rights.

The co-legislators also added a new provision banning AI practices involving the generation of non-consensual sexual or intimate content and child sexual abuse material.

Registration, data use and national “sandboxes”

Providers would again be required to register systems in the EU database for high-risk AI even when they consider their systems exempt from being classed as high-risk, the Council said.

The deal would also reinstate a “strict necessity” standard for processing special categories of personal data when it is used to detect and correct bias in AI systems.

National authorities would have until 2 August 2027 to set up AI regulatory sandboxes — controlled testing environments intended to help develop and trial AI under regulator oversight — under the provisional agreement.

Companies would have three months, rather than six, to put in place transparency measures for artificially generated content, with a new deadline of 2 December 2026.

The agreement also sets out how the EU’s AI Office would supervise some AI systems based on general-purpose AI models when the same provider develops both the model and the system, while listing areas where national authorities would remain responsible, including law enforcement, border management, judicial authorities and financial institutions.

The provisional deal must still be endorsed by both the Council and the European Parliament before legal and linguistic checks and formal adoption in the coming weeks.


Copyright © 2026 The Brussels Times. All Rights Reserved.