With 13 days to go before the EU AI Act’s August 2, 2026 transparency deadline, the law Brussels has already agreed to rewrite is still, technically, the law. The Digital Omnibus on AI was signed on 8 July 2026, but hasn’t yet appeared in the Official Journal. Until it does, Article 50 and its companion obligations remain fully binding on the original schedule.

That’s an awkward gap, and it’s the gap compliance teams are now sitting in.

The Omnibus itself moved quickly by Brussels standards: the Commission proposed it on 19 November 2025, negotiators reached political agreement on 7 May 2026, Parliament adopted the text on 16 June, and the Council gave final green light on 29 June before signature in July. Official Journal publication is expected this month. The delta between “signed” and “published” is procedural, but the legal consequences aren’t.

Once printed, the Omnibus defers the stand-alone high-risk regime from 2 August 2026 to 2 December 2027, and pushes embedded high-risk systems in medical devices, machinery and toys to 2 August 2028. National regulatory sandboxes slide to 2 August 2027. A new prohibition on AI systems generating CSAM or non-consensual intimate imagery lands, with technical safeguards required by 2 December 2026, per Freshfields’ read of the final text.

What doesn’t move: Article 50’s transparency obligations, the February 2025 prohibitions, and the GPAI rules that went live in August 2025. Machine-readable marking of AI-generated content applies from 2 August, with a three-month grace period for systems already on the market, ending 2 December 2026.

The lesson is one Brussels-watchers know well. A political deal to soften a rule isn’t the same as the rule being softened. Providers get to comply twice.

Sources

Sources