Ireland is about to settle one of the most consequential open questions in technology law: when an artificial-intelligence system causes harm, how does a plaintiff sue, and what must they prove? The answer taking shape is notable for what it is not. There will be no bespoke Irish AI-liability statute. Instead, AI harms will be litigated through modernised product-liability rules, anchored by the EU's revised Product Liability Directive (Directive (EU) 2024/2853), which Ireland must transpose into national law by 9 December 2026.
That path became the only path on the table once Brussels abandoned the alternative. On 28 September 2022 the European Commission proposed a standalone AI Liability Directive (AILD) to adapt fault-based civil-liability rules to AI — most controversially through a rebuttable presumption of causation and a right for claimants to demand disclosure of evidence about high-risk systems. After negotiations stalled, the Commission listed the file for withdrawal in its 2025 Work Programme and formally withdrew it in 2025, a status confirmed on the European Parliament's legislative train. Europe will not get a dedicated AI fault regime. It will get product liability.
The strongest case for the directive Brussels killed
The argument for the AILD was serious and deserves a fair hearing. AI systems are often opaque, probabilistic, and self-updating. A claimant injured by an automated decision or a faulty model may have no realistic way to reconstruct why the system behaved as it did, because the relevant logs, training data, and design choices sit entirely with the developer. Conventional negligence law asks the victim to prove duty, breach, causation and loss on the balance of probabilities — and causation is exactly where opacity bites hardest. The AILD's presumption of causality and disclosure right were designed to rebalance that asymmetry so that the hardest cases did not become unwinnable by default. That is a legitimate access-to-justice concern, not regulatory overreach.
The problem was duplication and uncertainty. By the time the AILD was debated, the AI Act had been adopted and the revised Product Liability Directive was already moving — both addressing overlapping ground. A coalition of industry bodies argued the AILD was premature, and the Parliament's own internal-market committee called its adoption at that stage unnecessary. Layering a third, fault-based instrument on top risked inconsistent national transpositions and litigation about which regime applied. Withdrawal was the proportionate call.
What actually changes in Irish courts
Ireland's current regime, the Liability for Defective Products Act 1991, transposed the original 1985 directive and is built around tangible products. As the Legal 500 Ireland AI guide notes, it "does not specifically refer to software or AI," so standalone software has fallen outside strict product liability unless embedded in a physical good. That gap closes in December 2026.
The revised directive does real work, and most of it is sensible:
- Software and AI are products. Strict, no-fault liability extends to standalone software and AI systems — not just AI baked into hardware.
- Defects can arise post-sale. Manufacturers can be liable where a product is made unsafe by a missing software update or an unaddressed cybersecurity vulnerability, as Irish firm William Fry explains.
- Disclosure and an eased burden of proof. Courts can order manufacturers to disclose relevant evidence, and the burden of proof is alleviated in technically complex cases.
That last point matters: the disclosure right and burden-of-proof relief the AILD would have created for fault claims now exist, in a narrower form, inside the strict-liability product regime. The victim-protection objective survives; the duplicative second statute does not.
Getting transposition right
The Department of Enterprise, Trade and Employment ran a public consultation on the directive and has told the Oireachtas it expects to meet the deadline. The substance of how Ireland transposes will decide whether the regime is proportionate or chilling.
Three principles should guide the drafting. First, no gold-plating: Ireland should transpose the directive's text without bolting on extra Irish-specific presumptions or lower thresholds that would make it an outlier and deter the developers Ireland has spent two decades attracting. Second, calibrate disclosure: the new disclosure power is the regime's sharpest tool, and courts need clear proportionality limits so it compels genuinely relevant evidence without becoming a fishing expedition into proprietary models. Third, preserve the negligence baseline: for the fault-based cases the AILD would have covered, Ireland's flexible common-law negligence — duty, breach, causation, loss — remains, and it can adapt case by case rather than by statute.
The outcome is, on balance, the right one for an open, innovation-friendly jurisdiction. Victims of defective AI get a clear, strict-liability route with real evidentiary help. Developers get one coherent regime instead of three overlapping ones. The remaining hard cases — autonomous, self-learning systems where causation genuinely defies attribution — are left to courts to work out incrementally. That is how the common law has always absorbed new technology: not by pre-emptively legislating every contingency, but by deciding real disputes on real facts. Ireland's job now is to transpose cleanly and let its judges do theirs.