The EU’s AI copyright question isn’t one question.
It’s three: what the EU AI Act requires of GPAI providers on data documentation, what the 2019 Copyright Directive allows on training data access, and what litigation brought by rights holders is testing in real time. These tracks are legally distinct. They share the same practitioners, the same datasets, and the same August 2 deadline pressure. Understanding how they interact is the work compliance teams in the European market need to do before summer ends.
Three Stakeholders, Three Positions
The policy debate has defined positions. CCIA Europe, an industry association representing major technology companies, launched “The TDM Equation” at the European AI Roundtable on Copyright on approximately June 3, projecting that restricting the EU’s text and data mining exception could cost the EU economy up to €600 billion annually, according to CCIA Europe’s commissioned study. The study was conducted by Implement Consulting Group and the Ifo Institute but commissioned by an industry group with a direct economic interest in the outcome. These figures have not been independently verified; cite them only with that attribution attached.
Content rights holders, publishers, news organizations, collecting societies, occupy the opposing position. Their argument isn’t primarily economic; it’s structural. The 2019 Copyright Directive’s opt-out mechanism for TDM was designed to give rights holders a practical way to exclude their works from AI training. Rights holders argue the mechanism hasn’t worked: opt-out signals are ignored at scale, and the burden of policing compliance falls on the rights holder rather than the trainer. The CNN v. Perplexity litigation (covered in four prior briefings from May 29–30 in this hub) and the Penguin Random House Munich case represent litigation-front enforcement of the same argument.
The European Commission occupies the middle position, actively reviewing whether to reopen the 2019 Directive’s TDM provisions, with no stated outcome yet. The Commission’s deliberation will weigh the industry’s economic projections against rights-holder documentation of opt-out mechanism failures. The €600 billion figure is now in that deliberation; whether it survives independent scrutiny determines its policy weight.
The GPAI Compliance Track Is Separate, and Intersecting
The EU AI Office reportedly published a GPAI FAQ addressing compliance questions for providers covered under Articles 51–56 of the EU AI Act, according to initial reporting, though the specific document URL hasn’t been independently confirmed for this package. The FAQ reportedly addresses how compute thresholds interact with systemic risk classification and what energy documentation obligations apply to GPAI model providers.
Timeline
Who This Affects
Eight weeks from publication, the August 2 deadline is close enough that “reportedly” isn’t a comfortable word for compliance teams to be operating with. The GPAI track and the copyright track interact specifically here: Article 53 of the EU AI Act requires GPAI providers to maintain a detailed summary of training data, including data sources used. If the training data includes material covered by the 2019 Copyright Directive’s TDM exception, that documentation obligation intersects with the TDM policy debate directly, a provider that relied on TDM-covered training data has created a documented record that rights holders and regulators can reference if the exception is subsequently challenged or restricted.
The open-source exemption adds another layer. The GPAI FAQ reportedly addresses whether releasing model weights publicly changes a provider’s documentation obligations. That question matters because a number of European AI developers have released models under open licenses, calculating that the compliance burden shifts. If the FAQ’s answer to that question narrows the exemption, requiring open-source GPAI providers to maintain Article 53 documentation even after public release, it changes the compliance calculus for a significant portion of the EU AI market.
What Litigation Is Testing
The copyright litigation pattern in this hub’s registry establishes that rights holders aren’t waiting for legislative clarity. Four briefs from May 29–30 document CNN’s suit against Perplexity AI, covering 17,000+ alleged works. The Penguin Random House Munich case tests similar theories in the German court system. These cases are proceeding under existing copyright law, the outcome doesn’t depend on whether the Commission reopens the 2019 Directive.
This is the fault line. The Commission’s TDM review is a legislative process with an uncertain timeline. Litigation is live now. A developer that relied on the TDM exception to train a GPAI model, and documented that reliance for Article 53 purposes, may face litigation exposure based on existing law before the Commission’s review concludes. The GPAI documentation requirement doesn’t create litigation exposure; it creates a detailed record of what happened. What that record is used for depends on how courts rule on existing TDM exception claims.
What European AI Developers Must Do
The action framing here is audience-specific. Two groups face distinct obligations on intersecting timelines:
*GPAI model providers (August 2 deadline):* Finalize Article 53 training data documentation now.Document which training data was sourced under TDM exception reliance and which was licensed directly, that distinction will matter if litigation reaches the data provenance question. Don’t assume the open-source exemption eliminates your documentation obligation until you’ve reviewed the FAQ’s specific language.
EU AI Copyright Compliance Risk, GPAI Providers
*AI developers deploying in the EU market (TDM policy track):* Monitor the Commission’s TDM review timeline closely. If your training pipeline relies on TDM-covered data and rights holders have signaled opt-out preferences you’ve honored, document that compliance now, before litigation asks for it. If your pipeline hasn’t been systematically honoring opt-out signals, the litigation pattern suggests that exposure is building across multiple jurisdictions simultaneously.
The real question is whether the August 2 GPAI deadline forces a documentation exercise that inadvertently maps litigation exposure. It might. But the alternative, failing to complete Article 53 documentation to avoid creating a record, trades certain regulatory violation for uncertain litigation exposure. That’s not a trade most EU-market AI developers should be making.
The Synthesis
CCIA Europe’s €600 billion figure will circulate in Commission deliberations regardless of its methodological rigor. Rights holders will challenge it with their own data. The Commission will weigh both. The litigation will proceed under existing law while that process runs. And GPAI providers will hit August 2 with documentation that establishes, in writing, exactly how they approached training data.
The fault line between these tracks isn’t going to close before summer. European AI developers should be operating on all three tracks simultaneously, not sequentially, and not by treating any one track as the definitive frame. The EU AI Act copyright question is three questions. The compliance program has to answer all three.