Source: Reed Smith

A date worth marking in the calendar: 2 August 2026 triggers the application of key parts of the EU AI Act, notably the transparency obligations under Article 50 covering AI that interacts with people, AI that generates synthetic content (Article 50(2)), emotion‑recognition/biometric categorisation (Article 50(3)), and deepfakes or text on matters of public interest (Article 50(4)). The Omnibus measure in force since 27 July 2026 alters the timeline: AI-generated or manipulated outputs falling within Article 50(2) and deepfakes within Article 50(4), first subparagraph, that were produced before 2 August 2026 do not need retroactive labeling. Enforcement formally begins for provisions already in force, including prohibited practices, GPAI obligations, the transparency rules, and AI literacy requirements.

Several significant deadlines were postponed by the Omnibus. The new prohibitions on non‑consensual deepfakes and CSAM‑generating AI have been deferred to 2 December 2026, matching the transition deadline for machine‑readable marking under Article 50(2) for systems already on the market. High‑risk obligations under Annex III (biometrics, critical infrastructure, education, employment, migration/asylum/border control) will not apply until 2 December 2027; Annex III’s substance is unchanged and no new high‑risk use‑cases were added. Under Annex I, only AI embedded in machinery‑regulated products is now excluded from the high‑risk regime, while other Annex I products (toys, lifts, medical devices) remain subject to compliance by 2 August 2028. The Omnibus also introduces a “small mid‑cap” definitional category (under 750 employees, under €150M turnover) granting SME‑level relief, and Member States’ obligation to operate at least one AI regulatory sandbox has moved from 2 August 2026 to 2 August 2027. The piece notes Kinza Yasar’s cited article reflecting an interview with the author, Dr. Andreas Splittgerber.