This is not a new EU AI law. That framing matters before anything else.
The European Parliament’s resolution 2025/2058(INI), passed on March 10, 2026, calls for the European Commission to act on AI copyright, mandatory transparency for training data, a compensation framework for rights holders, and provisions addressing AI use of content in retrieval-augmented generation systems. Parliamentary own-initiative resolutions of this type are non-binding instruments. They carry political weight and put formal pressure on the Commission to respond, but they create no direct legal obligations on AI companies or any other party.
What the resolution actually calls for
The resolution’s key demands cluster around three areas. First, mandatory transparency: AI companies would be required to publicly list the copyrighted works used in training datasets and maintain records of content used in data crawling. Second, a compensation framework: rights holders would receive structured rights to compensation for the use of their works in AI training. Third, RAG-specific provisions: the resolution addresses content used not just in training but in retrieval-augmented generation, where model outputs draw directly on indexed content at inference time.
Legal analysts at Lynx Legal have interpreted the resolution’s compensation provisions as potentially extending to past usage of copyrighted content, a retroactive compensation framework. This reading is contested; it reflects interpretive analysis of what the resolution implies, not text the resolution contains directly. The “no transparency, presumed infringement” shorthand circulating in legal analysis circles similarly reflects analyst characterization of the transparency requirements’ implications, not verbatim resolution language.
Why news media rights holders are the pivotal stakeholder
News media organizations have been among the most vocal advocates for RAG-specific provisions. Their position is distinct from general training data copyright concerns: they’re not just asking whether their archives can be used to train models, they’re asking whether their content can be retrieved and incorporated into AI-generated outputs at inference time, without licensing and without compensation. This distinction matters because RAG use is ongoing and cumulative in ways that one-time training data ingestion is not.
What to watch: the Commission’s summer window
The Commission is expected by legal analysts to address AI copyright in a forthcoming Digital Omnibus package, with a summer 2026 revision window targeted. The Commission has not confirmed this timeline. What the Commission includes, or doesn’t include, in that package will determine whether the Parliament’s resolution generates any binding legal framework, or whether the copyright question extends into 2027 without resolution.
For the US/UK comparison on how different jurisdictions have handled AI copyright, see the US vs. UK AI copyright analysis. For the White House’s contested position on training data safe harbors, see the White House AI framework copyright brief.
TJS synthesis
The resolution is a signal of political direction, not a change in legal requirements. For AI companies using EU-sourced training data or operating RAG systems that index EU content, the practical implication is monitoring, not immediate compliance action. The Commission’s summer window is the decision point. If the Digital Omnibus incorporates the Parliament’s transparency and compensation demands into binding text, the compliance landscape shifts materially. If the Commission doesn’t act, or acts in a narrower form, the Parliament’s resolution becomes a data point in an ongoing negotiation. Either way, the resolution itself is not the trigger.