fair use? A German Court Just Dismantled That Defense — With American Law

The most immediate landscape is dense with court deadlines. UMG filed its motion to dismiss on August 5, built on the argument that "A payment measured by an agreement that does not exist is no payment at all" and that the Sound Recording Labor Agreement "forecloses AFM's sole claim as a matter of law." Warner Records and Atlantic Recording have not yet filed ahead of their August 14 deadline, and briefing doesn't close until September 11. I think the motions to dismiss are more likely to be denied than granted — AFM's video game precedent, in which "new use" treatment was applied without any separate AFM agreement covering that medium, is a historically grounded counterargument that's genuinely difficult to set aside cleanly. If the court denies dismissal, the case moves into substantive proceedings and its industry-wide impact expands significantly. August 25 also marks the document production deadline in Sony v. Udio — the existing suit covering 333 songs — meaning multiple AI music legal fronts are converging in the same narrow timeframe.

The discovery phase, if AFM's case survives dismissal, may matter more than the eventual verdict. Discovery would compel labels to disclose the specific terms of their AI licensing agreements: deal amounts, scope of recordings covered, and what if any notifications reached the musicians whose recordings were included. All of this is currently confidential. If those terms enter the court record, it will be the first time the actual mechanics of label-AI deals are visible to the public, to other musicians, and to other unions. I believe that moment of forced transparency could generate more industry disruption than the ruling itself. The instant that information becomes accessible, other musicians and labor organizations will begin demanding equivalent disclosure — and the legal pressure will compound across the industry. Leaking that the label-AI deal terms were structured to systematically exclude musicians would be far more damaging than losing a motion.

The GEMA appeal is among the most important near-term variables to track. Since the ruling is non-final, an appeal from Suno is essentially expected. The appellate proceedings will most likely center on two issues: the extraterritorial application of U.S. law and the standard for what constitutes model-level memorization of protected works. I think the extraterritorial question will be the most fiercely contested. Munich's dual choice-of-law framework is legally coherent, but it represents a significant step — one court interpreting another nation's law in a binding judgment — and appellate courts will examine it with particular care. The interesting wrinkle is that the longer the GEMA appeal takes, the more likely similar proceedings in France, the UK, or other European markets reach conclusions first. Organizations like SACEM and PRS are watching Munich closely, and "if Munich can do it, so can we" is not an implausible next move for any of them.

Looking at the medium term, AI music companies' business models will likely require fundamental restructuring. IFPI's 2026 report puts global recorded music revenues at $31.7 billion, with 837 million paid streaming subscribers and $16.6 billion from paid streaming. Suno alone generates $300M in ARR — less than 1% of that $31.7 billion market. If the Munich ruling holds on appeal and spreads across European markets, every AI music service operating in Europe will need to establish formal licensing arrangements with national rights management organizations. The current cost structure of AI music generation assumes training data costs essentially nothing — zero acquisition cost for the raw material. Add licensing fees and the unit economics change fundamentally. Subscription prices would need to rise, user growth would slow, and the industry's growth trajectory would face structural headwinds. I don't think that's necessarily a bad outcome. If the growth to date was subsidized by zero-cost appropriation of others' creative work, a correction toward proper pricing represents a healthier equilibrium — provided the licensing fees actually flow through to musicians rather than being captured by labels.

The "new use" question will ripple far beyond music if AFM prevails on the merits. The core principle — that when technology advances beyond what a labor contract contemplated, workers deserve additional compensation for those new uses — applies directly to every creative industry labor agreement written before AI existed. SAG-AFTRA is already in active discussions over AI actors and digital replicas. The Writers Guild has raised AI learning compensation as a central demand. A legal win for AFM establishing that AI training qualifies as "new use" would give all of these organizations a directly applicable precedent to build on. I believe AFM's case could become the first formal legal test for redefining labor rights in the AI era — not just for musicians, but for every creative worker whose output serves as training material for commercial AI systems.

Streaming platforms' evolving policies add another layer to the medium-term picture. Spotify, Apple Music, and similar services are already adjusting their AI music distribution policies, and cases of platforms refusing to list AI music with unclear licensing are multiplying. Apple Music's own data is striking: 33% of new uploads are AI-generated, yet AI accounts for under 0.5% of listening time. Deezer mirrors this — 28% of uploads, 0.5% of streams. If Munich's ruling holds on appeal, platform-level gatekeeping will intensify further, creating a dual-market structure: a licensed premium tier with full distribution access, and a gray-zone tier operating outside legitimate channels. This mirrors the early streaming era dynamic between Napster and iTunes — and suggests that the legitimacy filter may arrive through market mechanisms even before any court issues a definitive ruling.

The long-term structural question is whether a global AI training data framework emerges that functions like GDPR did for data privacy. If Munich's dual choice-of-law approach is adopted across European jurisdictions, AI companies will need to satisfy the copyright laws of every nation whose music their training data includes — simultaneously. That is a dramatically more demanding standard than the current "train anywhere, defend later" posture. The EU AI Act's copyright-related provisions, as they come into force, combined with GEMA-style court rulings, could produce a "copyright GDPR effect" — where global AI companies ultimately apply the strictest standard uniformly, because managing market-by-market variation is operationally more costly than consistent compliance. This won't happen quickly. Copyright law varies far more dramatically between nations than data privacy law, and aligning international standards will require years of multilateral coordination. But the trajectory is becoming visible, and companies that ignore it are taking on compounding legal risk.

The most optimistic scenario I can construct: Munich's ruling holds on appeal, AFM wins substantive proceedings or extracts a structured settlement with mandatory musician compensation requirements, and a framework emerges requiring AI training revenue to be shared with rights holders in a transparent, auditable way. AI music companies operate under increased costs but gain legal certainty, and investment stabilizes on that foundation. The base case is more measured: Munich's appellate ruling modifies the first instance on extraterritoriality, AFM's case settles out of court for symbolic compensation without meaningful structural change, and labels retain dominant control of AI licensing revenue allocation — establishing principles on paper without changing who keeps the money. The pessimistic scenario: the appellate court overturns the extraterritorial application entirely, AFM's case is dismissed as outside the SRLA's scope, and labels win exclusive control of AI licensing, giving legal sanction to an arrangement that systematically excludes musicians from AI deal proceeds.

The variable most likely to make my forecast wrong is U.S. fair use jurisprudence developing in a direction favorable to AI training. If American courts conclude that AI training is broadly transformative under fair use doctrine, Munich's ruling becomes a European carve-out rather than a global signal. It's also possible that AI architectures evolve rapidly toward synthetic training data, public domain sources, or consent-based artist partnerships — in ways that make the licensing dispute over existing catalogs less central over time. But the retroactive liability question — what legal status applies to models already trained on unlicensed data — will not disappear regardless of which technical direction the field moves. That question, how the law handles AI systems built on data collected without clear consent, may turn out to be the hardest and most enduring issue in AI copyright law. No court has answered it definitively yet, and the answer, when it comes, will carry consequences far beyond the music industry.