The Digital Omnibus: The New EU AI Act Deadlines Explained

The EU AI Act's headline deadlines have moved. The Digital Omnibus on AI — provisionally agreed on 7 May 2026 and approved by the European Parliament in plenary on 16 June 2026 — defers the high-risk obligations and the fundamental rights impact assessment (FRIA) from 2 August 2026 to 2 December 2027 and 2 August 2028, adds a new prohibition on 'nudifier' and CSAM-generating tools, and simplifies several obligations. This guide explains what changed, what is still pending, and why the original dates remain the binding law for now.

"Two things are true at once: until the amendment is published in the Official Journal, the original AI Act dates remain the binding law; once it is published, the high-risk dates move to 2 December 2027 and 2 August 2028. Prepare for both."

How We Got Here

The Digital Omnibus is an amending regulation, not a new act. The European Commission tabled it on 19 November 2025 as part of a wider simplification package. Parliament and Council reached a provisional trilogue agreement on 7 May 2026, Coreper confirmed it on 13 May 2026, and the IMCO and LIBE committees cleared it on 2 June 2026. On 16 June 2026 the European Parliament approved the text at first reading by 423 votes to 57, with 174 abstentions.

The only open variable now is the exact Official Journal publication date. Until that publication, the original AI Act dates remain in force as a matter of law.

The Corrected Timeline

Here is the full phased timeline as it stands after the Digital Omnibus. The dates marked 'pending OJ' take legal effect only once the amendment is published.

What 2 August 2026 Now Means

This is the most important correction. The 2 August 2026 date was widely billed as 'full application' for high-risk AI. It is not. After the Digital Omnibus, 2 August 2026 is essentially the transparency date: Article 50 obligations apply, meaning people must be told when they are interacting with an AI system, and AI-generated or manipulated content (deepfakes, synthetic audio, images and text) must be marked as artificially generated.

The high-risk system obligations under Articles 8-15, the operator duties, conformity assessment, and the FRIA under Article 27 all move to the deferred dates below — they are no longer due in August 2026.

The New Prohibition: 'Nudifier' and CSAM Tools

The Digital Omnibus adds a new Article 5 prohibition, effective 2 December 2026, targeting AI systems designed to generate non-consensual intimate imagery ('nudifier' apps) and child sexual abuse material. The same date marks the end of the Article 50(2) grace period — the machine-readable marking of AI-generated content must be in place even for systems that were already on the market before 2 August 2026.

High-Risk and FRIA: Deferred to 2027–2028

The high-risk framework now applies on two new fixed dates. Stand-alone high-risk systems listed in Annex III — the use-case pathway, covering recruitment, credit, insurance, education, essential services and more — apply from 2 December 2027, a 16-month deferral. This is also the new FRIA deadline under Article 27. High-risk AI embedded in products covered by EU harmonised safety legislation (Annex I) — medical devices, machinery, vehicles, toys, lifts and similar — apply from 2 August 2028, a 12-month deferral.

"The FRIA tracks the Annex III high-risk applicability date. With that date now 2 December 2027, the FRIA deadline moves with it — but only once the amendment is published. Until then, 2 August 2026 remains the date in law."

Beyond the Dates: Other Substantive Changes

The Digital Omnibus is more than a timeline shift. The provisionally agreed text also makes several substantive changes:

• AI literacy (Article 4) is softened from a duty to 'ensure' literacy to one to 'take measures to support the development of' literacy
• The AI Office gains exclusive competence over AI systems built on a provider's own general-purpose AI model and over AI in very large online platforms and search engines, plus inspection, binding-commitment and fining powers
• The 'safety component' concept is narrowed — AI used only for assistance, optimisation, efficiency or convenience is not high-risk unless its failure endangers health or safety
• The Machinery Regulation moves to Annex I Section B, so sectoral safety rules take precedence over dual compliance
• A new Article 4a opens a narrow pathway to use special-category data for bias detection and correction, under strict safeguards
• Article 25 value-chain information duties are sharpened; breaches now sit in the 3% of worldwide turnover / EUR 15M fine band under Article 99(4)

What You Should Do Now

Sources

• European Commission press release IP/26/1024, 'EU agrees to simplify AI rules… ban nudification apps', 7 May 2026
• Council of the EU, 'Artificial Intelligence: Council and Parliament agree to simplify and streamline rules', 7 May 2026
• European Parliament, 'AI Act: deal on simplification measures, ban on nudifier apps'; plenary approval 16 June 2026
• Commission draft guidelines on classification of high-risk AI systems (Article 6), 19 May 2026
• Gibson Dunn, 'EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines', 27 May 2026
• Covington / Global Policy Watch, 'EU AI Act Update: Timeline Relief, Targeted Simplification, and New Prohibitions', 2 June 2026
• White & Case, 'EU agrees Digital Omnibus deal to simplify AI rules', 14 May 2026
• Council document ST 9247/2026 (final package); Bird & Bird consolidated AI Act + Digital Omnibus text

Key Takeaways