EU AI simplification package reaches critical milestone

The parliament’s plenary voted today in favour of proposals to finalise amendments of the AI Omnibus, which aims to simplify the implementation of the EU AI Act across the EU and reduce the administrative burden for businesses.

The AI Omnibus forms part of a wider ‘digital simplification’ package put forward by the European Commission in November 2025. While most provisions are due to take effect from 2 August 2026, EU ministers have been under pressure to finalise details, including specific timelines for different provisions to take effect to ensure member states and businesses have enough time to prepare.

On 13 March the EU Council agreed its general approach to streamlining certain rules within the package.

However, as the simplification package is subject to the usual EU law-making process, it required formal agreement between lawmakers in the European Parliament and Council of Ministers on the wording of reforms before they can be written into EU law and take effect.

On 18 March MEPs proposed to postpone the activation of certain rules on high-risk AI systems to give businesses a clearer timeline to adjust to the new requirements.

MEPs proposed that the activation of high-risk AI systems – including those involving biometrics, used in critical infrastructure, education, employment, essential services, law enforcement, justice or border management – be delayed until 2 December 2027.

MEPs also proposed to give businesses more time to comply with new rules for AI systems covered by EU sectoral legislation on safety and market surveillance, tabling amendments to postpone the activation date for these systems until 2 August 2028.

If approved, providers will also have until 2 November 2026 to adjust to new rules on watermarking AI-created audio, image, video or text content to indicate its origin. This stopped short of the Commission’s proposal to extend the deadline until 2 February 2027.

MEPs also proposed a ban on AI “nudifier” systems that use AI to create or manipulate images that are sexually explicit, intimate or resemble an identifiable real person without that person’s consent.

Today’s plenary endorsement – with MEPs voting in favour of the proposals by 569 votes – confirms and informs the parliament’s position going into the trilogue negotiations to finalise how the rules will be enforced and how the timelines – including for high-risk AI systems – might be adjusted.

The Council also proposed to postpone the deadline for establishing AI regulatory ‘sandboxes’ until 2 December 2027. According to the Commission, sandboxes will offer a “controlled environment” for providers to test their AI systems under supervision by national competent authorities and help mitigate the risk of non-compliance.

The Commission also clarified the mandate of the EU’s AI Office, confirming that it will be responsible for overseeing compliance of AI systems based on general-purpose AI models where the model and that system are both developed by the same provider, with notable exceptions for law enforcement border management, judicial authorities and certain financial institutions.

The simplification package aims to reduce the administrative burden of the EU AI Act on all businesses by at least 25% by 2029, and by 35% for small and medium enterprises (SMEs) in particular. However, Council has also called on the Commission to provide further guidance to help streamline the Act’s implementation to make it easier for businesses to get up to speed with the new rules, particularly with regard to high-risk AI systems.

Nils Rauer, a digital transformation expert with Pinsent Masons in Frankfurt, said agreement between lawmakers in the European Parliament and Council of Ministers on delaying the deadline for high-risk system rules coming into force would be a boon for businesses. “The industry has been calling for a delay in that deadline, as necessary standards and guidelines have been delayed,” he said. “The lawmakers have heard that call and both the Council and the Parliament are keen on delaying that applicability deadline, as both institutions have set new, fixed timelines.”

Today’s vote paves the way for the trilogue negotiations to start between representatives of the Council and the Parliament, with the Commission in a facilitating role. The first technical meetings are expected to take place later today and continue over the Easter period, underlining both the tight timeline and the political will to meet the August deadline.

Wesley Horion, an EU digital regulatory expert at Pinsent Masons, said the proposals were encouraging for the forthcoming negotiations. “Even before the formal trilogue negotiations start, we can already see some convergence happening between the Council's and the Parliament's respective positions,” he said. “For example, the Commission proposal to shift the onus of AI literacy from providers and deployers of AI systems to member states and the Commission has been pushed back against by the co-legislators. A parallel reading of the negotiating positions reveals interesting points of convergence and divergence before the start of the trilogue.”

Although the proposals provide businesses with some legal certainty, Rauer warned that businesses should not be complacent before they are formally adopted. “If for some reason the proposal was not to be agreed upon in time, the current deadline would still apply from 2 August 2026. Political processes are sometimes unpredictable and the talks may collapse on unexpected details; companies should therefore plan and prepare accordingly with the current timeline.”