GEMA v. OpenAI lawsuit update: Munich ruling is on appeal

GEMA v. OpenAI lawsuit update, last verified August 26, 2026: the Munich I Regional Court ruling dated November 11, 2025, case 42 O 14139/24, remains an influential EU first-instance decision on AI memorization and copyright reproduction. It is also on appeal. OpenAI’s appeal is pending before the Higher Regional Court Munich under 6 U 3662/25 e, and the cited May and June 2026 procedural sources report no appellate judgment or settlement [1][2][3].

The practical citation rule is therefore simple: cite GEMA v. OpenAI as the leading EU first-instance holding that memorized AI model material can amount to copyright reproduction, but disclose that it is not final. Do not describe it as settled German or EU law.

Source basis for this record: procedural posture is taken from AI Lawsuit Tracker’s May 2026 status page and Morgan Lewis’s June 25, 2026 German judicial-landscape update; the first-instance ruling is anchored in the Munich court’s November 11, 2025 press release and later English-translation access; the July 2026 Suno follow-on is treated separately because the available reporting is press-release-level rather than a full reasoned-judgment record [1][2][3][4].

The verified record to cite now

That table is the part most likely to matter in a brief, internal tracker, or background memo. The ruling is important because of what it says about model-level memorization, not because the appeal has already endorsed it. It has not.

What the Munich court actually held

The dispute concerned song lyrics administered by GEMA and outputs from OpenAI systems. The Munich court’s press release describes the case as involving lyrics from nine songs; this is the safer count to use because secondary reports have not been fully consistent on artist counts [3].

The model-level holding is the reason this case is being cited so heavily. The court treated memorized protected lyrics within the model as a legally relevant reproduction under §16 UrhG and Article 2 of the InfoSoc Directive. The careful version of the point is not that every act of AI training automatically creates an infringing copy of every work seen by a model. It is that, on the court’s findings, protected lyrics were memorized in a way that made the model’s stored representation legally meaningful as reproduction [3][5].

The output-level holding is separate. Where users prompted the system and received near-verbatim song lyrics, the court treated those outputs as reproductions and as making available to the public under §19a UrhG. For citation purposes, that distinction matters: one part of the judgment concerns what is stored in the model, and another concerns what the service delivers back to users [3][5].

The court also rejected the §44b UrhG text-and-data-mining defense. That point should be stated as a holding of this first-instance court, not as a final EU-wide answer to whether and when AI training can rely on TDM exceptions. The appeal, and potentially the Court of Justice of the European Union’s pending Like Company AG development discussed below, are exactly why the procedural label matters [5][2].

OpenAI also did not succeed in pushing legal responsibility entirely onto users. The Munich court imposed provider-side liability and granted GEMA injunctive relief, information/disclosure, and damages in principle, with the amount still to be determined. At the same time, it dismissed asserted personality-rights claims. A clean case note should include both sides of that relief posture; otherwise the ruling starts to look broader than it is [3][5].

The appeal posture is not a footnote

The pending appeal is not a technicality to bury after the merits. It is the difference between a strong citation and an overstatement. AI Lawsuit Tracker lists the case as appealed, and Morgan Lewis likewise reports the Higher Regional Court Munich appeal under 6 U 3662/25 e, with no appellate decision or settlement reported in the cited sources [1][2].

For current-status work, the tracker and the June 2026 law-firm update are the better sources for appeal posture. For the underlying ruling, use the Munich court release and translation-era analysis. Many November 2025 summaries were necessarily written from the German press release before wider access to the reasoned judgment. They can be useful for contemporaneous reaction, but they are a poor substitute for checking the later translation record before quoting the court’s reasoning.

If the case is going into a filing, the minimum disclosure should look something like this: “Munich I Regional Court, Nov. 11, 2025, 42 O 14139/24, appeal pending before Higher Regional Court Munich, 6 U 3662/25 e.” That one parenthetical prevents the common error: turning a live first-instance judgment into settled appellate law.

For teams maintaining AI copyright trackers, this is also a good candidate for a separate status flag rather than a generic “landmark ruling” entry. A broader monitoring workflow can sit alongside tools such as AI copyright and fair-use research trackers, but the case entry itself should preserve the appeal number and last-verification date.

2026 developments that change the weight of the ruling

April 2026: English translation access improved citability

The April 2026 translation milestone matters less as news and more as citation hygiene. The Max Planck Institute reported that a professional English translation of GEMA v. OpenAI was made free to view at GRUR International, after an AI translation had been published on April 4, 2026 [4].

That changes how English-language lawyers and journalists should handle the case. If a note still relies only on November 2025 commentary, it may miss the court’s reasoning or flatten the distinction between model storage, user-facing output, TDM, and remedies. The translation does not make the ruling final, but it does reduce the excuse for citing it vaguely.

July 31, 2026: the Suno follow-on points in the same direction, but is also non-final

On July 31, 2026, the same Munich I Regional Court chamber issued another GEMA copyright ruling, this time involving Suno, under case 42 O 763/25. Reed Smith’s report describes it as another GEMA win and notes two points likely to affect how observers read GEMA v. OpenAI: the chamber asserted jurisdiction over alleged U.S. training acts under §131 VGG, and it held that compliance with EU AI Act Article 53(1)(c)–(d) transparency obligations is not a defense to copyright infringement [6].

That is significant, but the source limitation should travel with the point. The Suno discussion available here is based on reporting from the court’s press release, not a full reasoned judgment, and the ruling is described as non-final with Suno able to appeal [6]. It is useful evidence that the 42nd Civil Chamber is extending or reinforcing related views; it is not an appellate confirmation of GEMA v. OpenAI.

September 3, 2026: Like Company AG is an expected citation-risk event

The CJEU’s Like Company AG opinion is expected on September 3, 2026. As of this August 26, 2026 status record, that is an expected date, not an issued opinion. Morgan Lewis flags the CJEU development in the same German AI-copyright landscape in which it discusses GEMA v. OpenAI [2].

That date should be treated as a re-check trigger. The CJEU opinion may confirm, narrow, or partially supersede reasoning that the Munich court used. Any filing, client memo, or publication after September 3, 2026 should verify whether the opinion issued and whether it affects reproduction, TDM, or liability analysis before leaning heavily on GEMA v. OpenAI.

Practical judgment for current use

GEMA v. OpenAI is important enough to cite because it squarely treats AI memorization of protected lyrics as copyright reproduction and pairs that with output-level liability and provider-side responsibility. It is too unsettled to overclaim because OpenAI’s appeal remains pending and no appellate ruling or settlement is confirmed in the cited status sources [1][2].

The disciplined citation is not “AI training is infringement under EU law.” It is: a Munich I Regional Court first-instance judgment held, on the facts before it, that memorized model material and near-verbatim outputs infringed German copyright law; the judgment is under appeal; related 2026 developments, including Suno and the expected Like Company AG opinion, should be checked before reliance.

References

1. GEMA v. OpenAI: Case Status May 2026, AI Lawsuit Tracker, May 13, 2026.
2. AI and Copyright – Judicial Landscape in Germany, Morgan Lewis, June 25, 2026.
3. LG München I press release, LG München I, November 11, 2025.
4. GEMA v OpenAI – Professional English Translation Free to View at GRUR International, Max Planck Institute.
5. GEMA vs. OpenAI: AI memorisation is a reproduction relevant to copyright law…, Osborne Clarke.
6. GEMA notches a second transatlantic AI copyright win in Germany, Reed Smith, July 31, 2026.