On 14 October 2026, at the Irish Film Institute in Dublin, Enterprise Minister Peter Burke closed the Irish Presidency conference "Copyright and Human Creativity in the Age of AI", which the Irish Recorded Music Association and NewsBrands Ireland organised. His speech set out a goal: "a thriving market for licensing of content for AI uses with fair remuneration along the value chain." It is the right goal. Reaching it depends on a few unglamorous fixes that the speech itself points to.
The strongest case for stricter rules
The rightsholders' argument deserves a fair statement. Musicians, publishers and journalists produce the material that makes generative models useful. They did not negotiate over its use, and they now compete with systems trained on their output. Burke reported that rightsholders are "struggling to enforce their rights – to declare an opt out in practice and to know if their works have been used in AI." If a creator cannot find out whether a work was ingested, the legal right to object has little practical value. Calls for new obligations, including mandatory remuneration, come from that gap.
What the law actually provides
The framework is the Copyright in the Digital Single Market Directive, Directive (EU) 2019/790. Article 4 allows text and data mining of lawfully accessible works unless the rightholder has reserved that use "in an appropriate manner." Recital 18 says that for content made publicly available online, reservation by machine-readable means, including metadata and a site's terms and conditions, is the appropriate route.
The design is a default permission with a right to opt out. That default lets European researchers and start-ups build models without negotiating thousands of licences first. The system's weakness is operational. "Machine-readable" was never tied to a single standard, so a creator who adds a robots.txt line or a metadata tag cannot be sure that every crawler will detect and honour it. The rightsholders' complaint in Dublin is largely about this practical gap, not about the principle of the exception.
Where the Commission is heading
Burke said the Commission is reviewing the Directive and "looks ahead to potential targeted measures, including in relation to generative AI." He described a process to assess the existing framework and, if needed, add EU-level measures to "encourage licensing, facilitate enforcement by rightsholders, and to address the challenge posed by deepfakes." He urged stakeholders to respond to the Commission's consultation by 3 November. That deadline fell after the 14 October speech and has now passed.
The Parliament's position is more interventionist. A Covington summary of the draft report on copyright and generative AI describes proposals for a central EUIPO opt-out register and a remuneration obligation for providers that aggregate press content. It also describes a demand for an "itemized list" of every protected work used in training. Leaked compromise amendments went further, suggesting that non-compliant providers could be barred from operating in the Union. These were committee-stage proposals, not adopted law.
Licensing works only if the plumbing works
Burke's phrasing is the more promising one because it points to a market, not a prohibition. Licensing markets form when buyers can identify what they are buying and sellers can enforce their terms. Three conditions follow.
- A usable opt-out standard. The EU should converge on a small set of interoperable machine-readable reservation protocols. A single clear signal protects creators and gives developers a compliance target. Today's ambiguity serves neither side.
- Proportionate transparency. Creators need enough information to know whether their works were used. A work-by-work inventory for every training run would be costly to produce and could hand a compliance burden to smaller European developers that the largest firms can absorb easily. A searchable, standardised disclosure of sources and crawlers' opt-out compliance would deliver most of the benefit at a fraction of the cost.
- Voluntary and collective licensing, not mandates. Collective management bodies can lower transaction costs for individual creators. Fixed statutory remuneration risks setting prices before anyone knows what training data is worth.
The free-expression and open-internet case matters here too. The text and data mining exception also supports search, archiving, research and accessibility tools. A regime that treats every automated read of a public web page as licensable would hit these uses as well as commercial model training. Targeted measures should be drafted around the specific problem, which is enforcement and information, not the act of machine reading.
Why Ireland's stake is unusually large
Ireland is a major European hub for AI and data-centre investment. Goodbody's November 2026 Economy Health Check found that AI and data-centre investment accounted for around one-fifth of growth in modified domestic demand in the four quarters to Q2 2026. Imports of AI and data-centre equipment used for capital investment reached €11.4bn, equivalent to 59pc of modified machinery and equipment investment. Ireland is also home to a creative and publishing sector that wants fair treatment. A Presidency that brokers licensing, not confrontation, serves both constituencies.
What to watch
The Commission's review will show whether it pursues interoperable opt-out standards and lightweight transparency, or follows Parliament toward registers, itemised lists and exclusion sanctions. The first path gives rightsholders enforceable rights and keeps Europe attractive to developers. The second risks driving training and deployment elsewhere without paying creators any more. Burke's emphasis on licensing, enforcement tools and fair remuneration suggests Dublin favours the first. The consultation has closed, so the test is whether the proposal that follows turns that emphasis into specific, workable mechanics.