The AI Omnibus entered into force on 27 July 2026. The main high-risk rules for Annex III systems now apply from 2 December 2027, while the corresponding Annex I product rules apply from 2 August 2028. These dates are enacted, rather than a political agreement awaiting adoption.
Which date applies to your system?
Start by establishing the system’s intended purpose and your organisation’s role. Annex III covers specified uses such as recruitment and certain other decisions affecting people. Annex I concerns specified regulated products and safety components. The relevant legal tests still determine whether the system is high-risk.
The new dates concern the provisions identified in amended Article 113. They are not a postponement of every obligation involving AI. Existing-system transitions, the particular article and any applicable exception need checking separately.
What still needs attention now
The original prohibited-practice provisions and AI literacy obligation began applying in February 2025. Article 4 has since been amended to require measures supporting literacy development, without mandating a particular individual level. Other law, including data protection and employment law where applicable, continues to matter.
Article 50 generally applies from 2 August 2026. Its duties differ: a provider’s technical marking obligation is different from a deployer’s disclosure duties for deepfakes or certain public-interest text. A single instruction to label every AI-assisted document is not an accurate account of the article.
Use the time to resolve specific gaps
Record the proposed or actual use and identify who can answer questions about it. Obtain the provider’s intended-purpose statement and instructions. Where high-risk duties will apply, work through oversight, data, monitoring and record-keeping with the people who will operate them.
Test whether an assigned reviewer can actually detect and act on a problem. If logs are needed, establish who holds them and whether your team can retrieve them. These practical dependencies are easier to fix before a deadline than during an assurance request.
Keep the classification under review
A supplier change, new use or significant alteration can change the analysis. Keep the reasoning and the source checked with the decision. An inventory supports that work, but an entry in software is not itself a legal conclusion or proof that controls operate.
AI Register can help organise systems, decisions and evidence. Use the current legal sources alongside the record, and obtain advice where the classification or your role is uncertain.
Related guidance: the Irish AI Act guide and recruitment AI classification.
Legal references checked 4 October 2026: AI Omnibus in force; Article 4 guidance; Current application dates; Article 50 transparency; High-risk classification; Irish Act and AI Office. Check the provision and any transition that applies to your particular system.
One specific transition matters: Article 111(4) gives providers of synthetic-content systems placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2). This is not a general postponement of deployer disclosures. Article 111 also contains separate existing-system high-risk transitions.