The Council has formally adopted changes delaying high-risk AI obligations and reducing regulatory overlap. The final vote gives companies clearer deadlines, but it also raises a harder question: whether simplification can preserve credible supervision of Europe’s AI rulebook.
The Council of the European Union has given final approval to a regulation simplifying parts of the AI Act, completing a rapid legislative revision before important high-risk obligations were due to begin applying.
The change gives companies and national authorities more time, but the political test is larger than the timetable. Brussels has promised both to cut administrative burden and to preserve the protections that made the AI Act a global regulatory model.
Formal adoption ends one uncertainty
EU Today covered the provisional institutional agreement in May and later examined the practical consequences of delaying key AI obligations.
The 29 June vote is nevertheless a material development. It converts the negotiated compromise into adopted law and provides fixed dates that compliance teams can use for planning.
Before the revision, application depended partly on whether standards, guidance and support tools were ready. Businesses argued that this created uncertainty: they could face legal obligations without the technical instruments needed to demonstrate compliance.
The new timetable removes some of that ambiguity. It also means the EU has accepted that its original implementation schedule was not realistic.
High-risk systems receive the longest delay
High-risk AI includes systems used in areas where errors or discrimination can seriously affect people, such as employment, education, critical infrastructure and access to important services.
Stand-alone systems will now face the relevant obligations from December 2027. Systems embedded in products governed by sectoral safety legislation, including some medical devices, toys, lifts and watercraft, will have until August 2028.
The additional time is intended to allow standards, guidance and supervisory structures to develop. Delay, however, also postpones enforceable protections in sectors where AI is already being deployed.
Regulators will need to use the transition period actively. If guidance arrives near the new deadlines, the same readiness crisis will simply reappear later.
Overlapping rules are being narrowed
The regulation creates a mechanism for cases in which sector-specific law imposes AI requirements similar to those in the AI Act. Implementing measures may limit duplicate application where equivalent obligations already exist.
That is potentially valuable for manufacturers. A company should not have to produce two sets of near-identical evidence for the same safety issue under two legal regimes.
The risk lies in deciding when requirements are genuinely equivalent. If overlap is interpreted too broadly, sectoral exemptions could create inconsistent protection. If interpreted too narrowly, the simplification promise will have little practical effect.
The Commission will be required to provide guidance for companies operating high-risk systems under harmonised product law.
The AI Office receives clearer authority
The text clarifies the EU AI Office’s supervisory competence where a general-purpose AI provider also develops a system based on its own model.
National authorities retain roles in sensitive areas including law enforcement, border management, judicial activity and financial institutions. This division is designed to reduce governance fragmentation while respecting sector-specific oversight.
It may not eliminate institutional disputes. Powerful AI systems cross legal categories and national borders, making responsibility difficult to assign when a model, provider, deployer and affected person are located in different jurisdictions.
The effectiveness of the reform will depend on information sharing and the willingness of authorities to act jointly rather than defend institutional territory.
New prohibitions accompany simplification
The regulation is not solely deregulatory. It adds prohibitions targeting AI-generated non-consensual sexual and intimate material and child sexual abuse content.
Systems that generate nude images of real people or digitally remove clothing to create intimate images are due to be prohibited from December 2026. The addition responds to the rapid spread of generative tools that can produce abusive material at scale.
The deadline for transparency tools covering artificial content is also set for 2 December 2026, with a shorter grace period for providers.
Those provisions allow Brussels to argue that simplification can coexist with new safeguards. Enforcement will determine whether the prohibition produces rapid removal, accountability and effective remedies for victims.
Competitiveness is now shaping implementation
The revision forms part of the Omnibus VII package and the EU’s wider effort to reduce regulatory costs. It follows repeated warnings that European technology companies face slower investment, fragmented supervision and heavy compliance demands compared with US and Chinese competitors.
Simplification can improve competitiveness when it removes duplication and uncertainty. It becomes deregulation when obligations are delayed without building the standards, staff and enforcement capacity required later.
The final vote therefore starts a new countdown. Companies have more time, but so do the Commission, AI Office, standards bodies and national regulators. Their work during that period will show whether the delay was used to make the system operational or merely to postpone conflict.
Brussels has now delivered the legal certainty industry requested. Its next obligation is to demonstrate that certainty does not come at the cost of credible oversight.
