Irish law for enforcing the EU AI Act set to be finalised

Though the AI Act is an EU regulation that has direct effect in EU member states, supplementary national legislation is required to give effect to some of the regulation’s provisions, including around enforcement. The Regulation of Artificial Intelligence Bill 2026 is the relevant legislation serving this purpose for Ireland.

The Bill provides for the establishment of a new AI Office of Ireland. The Office’s primary function will be to oversee compliance with the AI Act in Ireland. The draft legislation provides for enforcement processes and procedures, including scope to serve fines on businesses for non-compliance.

The Office will have other functions too – including a duty to promote and foster AI innovation and literacy – and will have powers to set up regulatory sandboxes to facilitate the testing of AI-related innovations by businesses. Another feature of the draft legislation is a mechanism for the reporting of serious incidents involving AI systems.

The Bill is proceeding through Ireland’s parliament, but it has already passed lower house without amendment, paving the way for it to be enacted by the upper house before the summer recess starts on 17 July.

The AI Act, dubbed the world’s first AI law, was written into EU law in 2024. The legislation sets out a risk-based approach to AI regulation, with some types and uses of AI prohibited altogether. The strictest regulatory requirements are otherwise reserved for AI systems that are classed as ‘high-risk’ AI systems and for certain ‘general purpose’ AI (GPAI) models.

The rules applicable to GPAI models took effect on 2 August 2025. Those rules require providers of those models to, among other things, produce technical documentation detailing the model’s training and testing process, maintain further information that allows others to integrate their models into their own AI systems, and make certain disclosures pertaining to the rights of copyright holders. A code of practice has been developed by the European Commission to support compliance.

The ‘high-risk’ AI regime was due to take effect on 2 August 2026, but EU law makers recently finalised legislation that will push that implementation date back – to late 2027 or mid-2028, depending on the type of system.

Providers and deployers of ‘high-risk’ systems face regulatory obligations around registration, quality management, monitoring, record-keeping, and incident reporting, while the systems themselves will need to requirements around risk management, data quality, transparency, human oversight and accuracy. Additional duties fall on importers and distributors. Pinsent Masons has developed a guide to help businesses understand whether the AI systems they develop or use constitute high-risk AI. The European Commission has also produced guidelines to support with self-assessments.