A date worth marking in the calendar: 2 August 2026 is a key milestone for the EU AI Act, though it is not quite the case that "the whole Act applies now." Part of it does. Another part, following the AI Omnibus, now in force since 27 July 2026, has been pushed further down the calendar. Here is a short overview.
What applies from 2 August 2026:
First, the transparency rules under Art. 50 start to apply. These cover AI that interacts directly with people (Art. 50(1)), AI that generates synthetic content (Art. 50(2)), emotion recognition/biometric categorisation systems (Art. 50(3)), and deepfakes or text on matters of public interest (Art. 50(4)). AI-generated or manipulated outputs falling within the scope of Article 50(2) AI Act and deep fakes within the scope of Article 50(4), first subparagraph, AI Act, which have been generated or manipulated before 2 August 2026 do not need to be marked or labelled retroactively.
Second, enforcement formally begins for everything already in force: prohibited practices, GPAI obligations, the transparency rules just mentioned, and AI literacy.
What does not apply yet – and this is where the Omnibus changed things:
The new prohibitions on non-consensual deepfakes and CSAM-generating AI have been postponed to 2 December 2026, the same date as the transition deadline for machine-readable marking under Art. 50(2) for systems already on the market.
The more significant shift concerns high-risk rules under Annex III (biometrics, critical infrastructure, education, employment, migration/asylum/border control), which were originally due to apply from 2 August 2026 as well. They now apply from 2 December 2027.
The substance of Annex III itself is unchanged. No new high-risk use-case was added. Under Annex I, only AI embedded in machinery-regulation products is now excluded from the high-risk regime, not other Annex I products like toys, lifts and medical devices, which still need to comply by 2 August 2028.
The Omnibus also adds a "small mid-cap" category (under 750 employees, under €150M turnover) with SME-level relief — a definitional change, not a new risk category.
Member States have been given more time as well: the obligation to have at least one operational AI regulatory sandbox has moved from 2 August 2026 to 2 August 2027.
In short: 2 August 2026 remains a genuine deadline, though not the finish line it was originally intended to be.
See some more details in the cited article by Kinza Yasar below that reflects a good portion of an interview with me.