The Digital Omnibus moved the high-risk deadlines — but three things still land in August 2026: AI literacy enforcement, Article 50 transparency, and the AI Office of Ireland. The plain-English timeline for Irish organisations.
If you have read a headline in the past month saying the EU AI Act has been "delayed", you have read something half true. The Digital Omnibus on AI — politically agreed on 6–7 May 2026 and confirmed by the Council on 13 May 2026 — moved the high-risk compliance deadlines back by sixteen months and two years respectively. It did not move anything else.
That distinction matters, because anyone still selling "high-risk panic before August 2026" is now factually wrong — and anyone telling you August 2026 no longer matters is wrong in the other direction. Three obligations genuinely land that month, and almost nobody is talking about the most enforceable of them.
Here is the timeline as it actually stands, in plain English.
What are the EU AI Act deadlines now?
| Date | What applies | Status after the Omnibus | | --- | --- | --- | | 2 February 2025 | Prohibited practices (Article 5) banned; AI literacy obligation (Article 4) applies | In force — unchanged (Omnibus adds new prohibitions on AI-generated NCII and CSAM) | | 2 August 2025 | GPAI provider obligations (transparency, copyright summaries) | In force — unchanged | | 1 August 2026 | AI Office of Ireland must be operational | Unchanged — set by Irish law, not the Omnibus | | 2 August 2026 | Supervision and enforcement of AI literacy begin; Article 50 transparency rules apply; Commission enforcement powers over GPAI providers activate | Unchanged | | 2 December 2026 | Grace period ends for machine-readable marking of AI-generated content by existing systems | Set by the Omnibus | | 2 December 2027 | High-risk obligations for stand-alone (Annex III) AI systems | Moved — was 2 August 2026 | | 2 August 2028 | High-risk obligations for AI embedded in regulated products (Annex I) | Moved — was 2 August 2026 |
One footnote before the detail: the Omnibus is politically agreed but not yet formally adopted. Official Journal publication is expected before 2 August 2026; until then, the original dates technically remain law. That is a residual-risk nuance worth knowing — not a reason to panic, and not a reason for anyone to keep selling the old deadline.
What moved?
The heavy machinery of the Act — conformity assessments, technical documentation, registration, human oversight architecture for high-risk AI systems — moved.
- Annex III stand-alone high-risk systems (recruitment and HR AI, credit scoring, AI in essential services, education, and similar): obligations now apply from 2 December 2027.
- Annex I embedded high-risk AI (AI in machinery, medical devices, vehicles, toys): obligations now apply from 2 August 2028.
The reason is practical rather than political: the harmonised technical standards organisations need in order to demonstrate conformity are not finished. The Omnibus also softened one piece of wording that matters for everyone — Article 4 now requires organisations to "support the development of" AI literacy among staff rather than "ensure" it. A real change in standard; not a repeal.
What didn't move?
Three things still land in August 2026, and they apply to far more Irish organisations than the high-risk rules ever did.
1. AI literacy enforcement begins on 2 August 2026
The Article 4 AI literacy obligation has applied since February 2025. What changes in August is enforcement: per the European Commission's own AI Literacy Q&A, supervision and enforcement of the literacy obligation begin on 2 August 2026.
This is the obligation almost nobody leads with, and it is the one with the broadest reach. It applies to every organisation that uses AI in any form — including Copilot, ChatGPT, and AI features embedded in standard business software. From August, a competent authority can ask an Irish organisation to show how it supports AI literacy among staff: who was trained, on what, calibrated to which roles and risks, with what records.
If your answer is "we sent around an e-learning module", you do not have an answer.
2. Article 50 transparency rules apply from 2 August 2026
From the same date, people must be told when they are interacting with an AI system, and AI-generated or manipulated content — deepfakes included — must be labelled. There is a grace period to 2 December 2026 for machine-readable marking by existing systems, but the disclosure obligations themselves proceed on schedule.
Any Irish organisation running customer-facing chatbots, AI-generated marketing content, or synthetic media needs these disclosures operating in production — not in a policy document.
3. The AI Office of Ireland opens on 1 August 2026
Ireland's enforcement architecture arrives regardless of the Omnibus. The General Scheme of the Regulation of AI Bill 2026 requires the AI Office of Ireland (Oifig Intleachta Shaorga na hÉireann) to be operational by 1 August 2026, coordinating a distributed model of 15 sectoral competent authorities — the Central Bank, the DPC, the HPRA, the CCPC, ComReg, the WRC, and the LSRA among them, designated in September 2025 — plus nine fundamental-rights authorities. The Commission's enforcement powers over general-purpose AI providers also activate on 2 August.
For most regulated organisations, the practical meaning is simple: your existing regulator becomes your AI supervisor, with a national coordinator behind it.
What does this mean for your organisation?
The honest answer is that the Omnibus changed the sequencing, not the work.
If you use AI at all — and the threshold includes mainstream office tools — the August 2026 items apply to you. AI literacy support, documented; transparency disclosures, where you deploy customer-facing AI; and a governance position your regulator can inspect.
If you deploy high-risk AI — recruitment screening, credit decisioning, clinical AI — you now have until December 2027 (or August 2028 for embedded systems) for conformity work. That is breathing room, not a reprieve. Conformity assessments, documentation, and oversight mechanisms take months to build, and the inventory and classification work that precedes them is also what the August 2026 obligations require.
If you are doing nothing, the demand-side data suggests you have company — and that is not reassuring. An IoD Ireland snap poll (February–March 2026, n=378) found 65% of directors don't understand what the new rules mean for their organisation, and only 48% of boards discussed AI governance in the past year. A William Fry survey reported by RTÉ in May 2026 found 55% of large Irish businesses unaware which EU laws apply to them. Mason Hayes & Curran's research found 67% of employers have no formal AI policy. The organisations that close those gaps before August will be answering routine supervisory questions; the rest will be explaining themselves.
What should you do before August 2026?
Four things, in order:
- Inventory your AI. Every tool, including the embedded and shadow-adopted ones. You cannot evidence literacy or transparency for systems you have not catalogued.
- Close the literacy gap — with evidence. Role-calibrated training, attendance records, a policy, and a documented rationale. The AI literacy obligation is the first thing enforced, and it is documentation-led.
- Check your transparency surface. Anywhere a customer interacts with AI or sees AI-generated content, the Article 50 disclosures need to be live by 2 August.
- Schedule the high-risk work against the real deadlines. December 2027 arrives faster than it sounds. A structured EU AI Act readiness review sequences the conformity work properly instead of hurriedly.
The full picture of Ireland's enforcement architecture — the AI Office, the 15 competent authorities, and what each can do — is covered in our guide to Ireland's AI regulation in 2026.
The Omnibus rewarded the organisations that ignored the panic and punished nobody. But it did not change the most basic fact of the EU AI Act era: from this August, AI governance in Ireland has a regulator, and "we hadn't started" stops being an acceptable answer.