EU Lawmakers Reach Provisional Agreement to Delay Key EU AI Act Obligations

Background

On 7 May 2026, following extensive negotiations, the European Council and European Parliament reached a provisional agreement on the EU Digital Omnibus on AI (AI Omnibus) which proposes targeted amendments to the EU Artificial Intelligence Act (AI Act). On 16 June 2026, the European Parliament voted to adopt the provisional agreement — although, formal adoption remains subject to European Council approval.

The EU AI Act is the EU’s flagship framework for governing the development, deployment, and use of certain AI systems in the EU. Although adopted in 2024, the regime was designed to apply on a staggered time line beginning in August 2025, with most obligations due to take effect on 2 August 2026. This included the core obligations for certain “high-risk” AI systems.

Given repeated concerns raised by industry and other stakeholders regarding, among other things, delays in the publication of standards and the associated compliance costs, the European Commission published the AI Omnibus on 19 November 2025. Through this proposal, the Commission aimed to introduce “simplification measures to ensure timely, smooth, and proportionate implementation of certain of the AI Act’s provisions.” Central to the proposal was a delay to the implementation timetable for “high-risk” AI systems.

Provisional Deal: Key Provisions and Time Lines

The provisional agreement — which marks a significant legislative breakthrough and gives organisations greater clarity on (amongst other things) the future compliance timetable for “high-risk” AI systems — must still be formally adopted to take effect. The European Parliament indicated in its press release that the intention is to do so before 2 August 2026, i.e., the current implementation date for the AI Act’s “high-risk” AI system rules.

Certain of the notable proposed amendments are set out below.

A. Delayed Application of Obligations for “High Risk” AI Systems

The application dates for obligations relating to “high-risk” AI systems will be amended as follows:

• 2 December 2027: rules for stand-alone “high-risk” AI systems (contained in Annex III of the AI Act), e.g., AI systems used in HR and employment.
• 2 August 2028: rules for “high-risk” AI systems embedded in products regulated under sector-specific legislation (contained in Annex I of the AI Act), e.g., AI-enabled medical devices regulated under the Medical Devices Regulation.

B. Delay to Watermarking Obligations for AI-Generated Content

The AI Act requires providers of AI systems generating synthetic audio, image, video, or text content to ensure such AI-generated content is identifiable through watermarking or similar technical measures. The application date for this obligation will be delayed until 2 December 2026, for providers who have already placed their systems on the market before 2 August 2026.

C. Reinstatement of Registration Obligations for Exempt “High Risk” AI Systems

The AI Act requires that certain “high-risk” AI systems are registered in an EU database. The AI Omnibus proposed removing this obligation where a provider concluded that the AI system was not “high-risk” by virtue of its reliance on an exemption under Article 6(3), e.g., where the AI system is intended to perform a narrow procedural task. However, this proposal has been dropped. As a result, providers relying on an Article 6(3) exemption will still be required to register the relevant AI system, although they will be able to do so through a simplified registration procedure.

D. Narrowing the Scope of “High-Risk” AI Systems Embedded in Products

Under Article 6(1), an AI system is classified as “high-risk” where: (i) it is intended to function as a safety component of a product or is itself regulated under Annex I legislation; and (ii) the safety component / product is required to undergo a third-party conformity assessment under that legislation.

The provisional agreement narrows what qualifies as a “safety component” such that products with AI functions that only assist users or optimise performance will not automatically face “high-risk” obligations, if their failure or malfunction does not create health or safety risks.

E. Ban on AI-Generated NCIC and CSAM

Parliament and Council have each confirmed the introduction of a ban on AI systems that generate child sexual abuse material (CSAM) or non-consensual intimate content (NCIC). This new prohibition will apply from 2 December 2026.

F. AI Literacy (see previous blog post, here)

 The provisional agreement shifts the obligation imposed on providers and deployers of AI systems to ensure AI literacy to one where the Commission and Member States “support and facilitate” providers / deployers to provision the development of AI literacy.

G. Other Notable Amendments

The provisional agreement also includes amendments addressing the following:

• Delaying the deadline for establishing AI regulatory sandboxes until 2 August 2027.
• Reinstating the “strict necessity” threshold for processing special category personal data for bias detection and correction purposes (i.e., ignoring proposals in the AI Omnibus and reverting to the status quo).
• Removing overlapping requirements for AI-enabled machinery products by clarifying that they only need to comply with applicable sectoral safety rules, rather than both the AI Act and sectoral rules, while ensuring an equivalent level of health and safety protection through appropriate safeguards to be proposed by the Commission.
• Extending certain SME exemptions to small mid-cap companies.
• Confirming that the EU AI Office will supervise AI systems based on general-purpose AI models developed by the same provider, while national authorities retain competence in specific sectors, such as law enforcement, border management, judicial authorities, and financial institutions.

Next Steps

As stated above, the provisional agreement (available here) remains subject to formal approval by the European Council. Once adopted, the legislation will be published in the Official Journal of the European Union and will enter into force three days later.

The European Council has indicated that formal adoption is expected “in the coming weeks.” Both co-legislators have also emphasised their intention to complete the process before 2 August 2026, i.e., the date on which the current “high-risk” AI obligations are otherwise scheduled to apply.

The European Parliament is simultaneously continuing work on the wider Digital Omnibus package, meaning further amendments to EU digital legislation are likely to follow.

Trainee solicitor Jennifer Petch also contributed to this blog post.

This post is as of the posting date stated above. Sidley Austin LLP assumes no duty to update this post or post about any subsequent developments having a bearing on this post.