Source: Taylor Wessing

On 6 May 2026, chief negotiators of the Council of the European Union and the European Parliament agreed on a provisional draft of the "Digital Omnibus on AI" to simplify the digital legal framework; the draft introduces significant changes affecting physical and industrial AI, including autonomous machines and robots. New application dates are set: rules for stand‑alone high‑risk AI systems apply from 2 December 2027, while high‑risk AI embedded in products apply from 2 August 2028. Providers will be required to register their AI systems in the EU database for high‑risk systems even if they do not consider them high‑risk themselves, and processing of special categories of personal data for detecting and correcting bias will only be permitted when absolutely necessary.

National authorities must establish AI regulatory sandboxes by 2 August 2027, and providers must implement transparency solutions for synthetically generated content within three months (deadline 2 December 2026). The Office for Artificial Intelligence’s responsibilities regarding systems based on general‑purpose AI models are clarified, although national authorities will remain responsible in certain cases. "Nudifier" apps that generate non‑consensual sexual or intimate content and depictions of child sexual abuse are to be banned under Article 5, with companies required to adapt by 2 December 2026. To avoid overlap with the EU Machinery Regulation (EU 2023/1230), machinery is moved largely out of the direct scope of the AI Regulation and the Commission is empowered to adopt delegated acts amending Annex III of the Machinery Regulation, to enter into force by 2 August 2028. During the transition, manufacturers may rely on harmonised standards or common specifications, but must conduct risk assessments based on current technical standards and the state of science and technology to mitigate liability risks; the briefing warns of greater short‑term responsibility, less double regulation medium‑term, and potentially stricter liability expectations.