Policy
EU Reaches AI Omnibus Deal — High-Risk Obligations Delayed, Machinery Exempted
European Parliament and Council negotiators struck a final compromise on the AI Omnibus in May, delaying high-risk AI obligations and removing machinery products from the AI Act's scope. Civil society groups warn the deal waters down fundamental safeguards.
By Michael C ·
Negotiators from the European Parliament and Council reached a final compromise on the AI Omnibus in a trilogue meeting held on May 6, 2026, resolving a months-long impasse over the scope of the EU AI Act's obligations. The deal, which the European Parliament aims to ratify in a plenary session between June 14 and 17, is designed to enter into force before the AI Act's high-risk obligations were originally scheduled to apply on August 2, 2026.
The central compromise keeps the AI Act's horizontal logic largely intact while making targeted concessions to industry. The most significant change is the exclusion of products governed by machinery regulation from the AI Act's scope — a long-standing demand from European manufacturers who argued that dual compliance requirements would impose disproportionate burdens on industrial automation companies. All other regulated product categories remain within the Act's scope, a point that drew criticism from industry groups who had sought broader exemptions.
What the Deal Changes
Beyond the machinery exemption, the final agreement includes several consequential changes that received less public attention. High-risk AI obligations will be delayed, giving companies additional time to prepare compliance programmes. The deal also allows the processing of special category data — including health, biometric, and political data — for debiasing purposes by a wider range of stakeholders and AI models than the original text permitted. Small and medium-sized enterprise exemptions have been extended to small mid-cap companies, and the powers of fundamental rights authorities to investigate AI systems have been circumscribed.
The AI Omnibus keeps the AI Act's horizontal logic intact — but delays high-risk obligations, exempts machinery, and narrows the powers of fundamental rights authorities. Critics say the deal trades safety for simplification.
A new provision banning AI systems that generate non-consensual intimate imagery or child sexual abuse material was included in the final text. However, the scope of the former ban — confined in a recital to depictions of listed body parts — has been criticised as unclear and potentially difficult to enforce in practice.
Mixed Reception from Stakeholders
Civil society organisations, while acknowledging that some of the most dangerous proposed changes had been avoided, reiterated concerns about procedural shortcomings — particularly the absence of an impact assessment for the omnibus amendments. The Centre for Democracy and Technology's European office argued that the AI Omnibus 'overall waters down safeguards against AI-related harm and fails to deliver on its promised simplification.' Other digital rights groups echoed this assessment, noting that the delayed entry into application of high-risk obligations effectively gives companies more time to deploy systems that may pose significant risks before facing regulatory scrutiny.
Industry actors welcomed the machinery exclusion but remained critical of the overall compromise. Business groups argued that the final law falls short of providing the clarity and simplification that European companies need to compete with American and Chinese AI developers operating under less stringent regulatory frameworks. Several major technology companies called on the EU to continue pursuing what they described as 'an ambitious simplification agenda' in future legislative cycles.
Implementation Guidance Still Pending
The European Commission has published two sets of draft guidelines to assist with AI Act implementation. The first addresses the classification of high-risk AI systems and is open for public consultation until June 23. The second covers transparency obligations under Article 50, including how operators must notify individuals that they are interacting with an AI system, how to mark AI-generated output, and how to disclose deepfakes. Stakeholders have until June 3 to provide feedback on the transparency guidelines.
The AI Omnibus represents the EU's attempt to balance its position as the world's most ambitious AI regulator with the economic imperative of keeping European companies competitive. Whether the compromise achieves that balance will become clearer as the implementation period unfolds. The delayed high-risk obligations and the machinery exemption suggest that the political pressure from industry was sufficient to soften the original framework — a pattern that critics argue will continue as AI regulation moves from legislation to enforcement.
Topics: EU, AI Act, Regulation, AI Omnibus, European Parliament