AI: Learning or Replicating?—What the Suno Ruling Asks of Japanese Music and Generative AI

By simply inputting text, a complete song with lyrics, melody, accompaniment, and vocals can be created in minutes. Music-generating AI is opening the door to music production for those without specialized composition skills, significantly transforming advertising, videos, games, and personal creative activities.

On the other hand, what AI listens to, what it remembers, and how it creates new music is mostly invisible to users.

On July 31, 2026, a ruling by the Munich District Court I in Germany shed judicial light on this opaque mechanism. In a lawsuit filed by the German music copyright management organization GEMA against the American AI music company Suno, the court largely upheld GEMA's claims for injunction, information disclosure, and damages.

The original article reports that a member of the German Social Democratic Party welcomed this decision, emphasizing the need to advance technological innovation while respecting the rights of creators. However, the significance of this ruling cannot be encapsulated in the simple statement "AI company lost in court."

The issue at hand was whether the generative AI merely learned general musical characteristics from copyrighted works or whether it retained specific works in a reproducible form within its model. This question is directly relevant not only to Germany but also to Japan, where the use of AI is rapidly expanding.

The Trial Was Not Just About "Resemblance to Famous Songs"

The trial focused on six musical works: the refrains of "Atemlos durch die Nacht," "Rasputin," "Big in Japan," "Forever Young," "Mambo No. 5," and "Daddy Cool." The lawsuit did not specifically address the infringement of lyrics rights but primarily concerned melody, harmony, rhythm, and song structure.

GEMA used Suno's AI model, inputting the original song's lyrics, title, and desired music style. However, they did not provide detailed instructions regarding melody, chord progression, rhythm, or arrangement.

Nevertheless, the generated music exhibited recognizable creative features of the original works. The court found it unlikely that complex, lengthy songs could coincidentally be reproduced in a similar manner.

A key issue here was "memorization," or the retention of training data by the AI model.

Generative AI companies generally explain that they do not store the learned works themselves like a database but rather learn characteristics and patterns from vast amounts of data in numerical form. Suno also argued that the model's weights and parameters only reflect generalized musical patterns, not the original songs themselves.

However, the court considered that if a specific work could be extracted from the model in a highly reproducible form, it goes beyond mere abstract learning, indicating that the content of the work is incorporated within the model.

In other words, the explanation that "it's not a reproduction because it's not stored as a file" was deemed insufficient.

The Responsibility for AI Outputs Was Not Placed Solely on Users

Another important point was recognizing the responsibility of the company providing the AI service, rather than the users, for the generated music.

Suno argued that the problematic audio was created from intentional input by GEMA, and that the user's actions led to the result. However, the court noted that the input instructions were relatively simple and did not specify the melody or arrangement in detail.

It is the service provider that selects the training data, designs the model structure, and operates the model. If the model memorizes specific works and generates outputs with those characteristics, it is the AI company that essentially determines the content of the results.

This perspective could significantly impact the business structure of generative AI services.

Previously, some AI companies took the stance that if problematic outputs occurred, the responsibility lay with users who violated prohibitions or input special instructions. However, this ruling indicates that if users can obtain outputs similar to existing works with simple operations, service providers cannot distance themselves by claiming "it was made by the user."

The Method of Obtaining Training Data Was Also an Issue

According to the court's official announcement, Suno's training data included the targeted songs, and the company was found to have used stream ripping technology to extract and reproduce music from YouTube. It was also explained that they bypassed technical measures designed to prevent downloads.

This is not simply a matter of "letting AI listen to publicly available music."

There is a distinction between being able to play music on the internet and an AI development company reproducing it on a large scale for use in commercial model training. Public content is not unconditionally reusable; terms of use, copyright, and technical protection measures are intricately involved.

For AI companies, it becomes crucial not only to demonstrate the performance of the completed model but also to explain where and how they obtained the training data.

Why the German Social Democratic Party Welcomed the Ruling

After the ruling, Martin Rabanus and Helge Lindh, members of the German Social Democratic Party, praised the decision as an important signal for creators and a fair digital market.

The emphasis was not on banning AI. They acknowledged that generative AI expands cultural and technological possibilities, but argued that the business should not be built on the unauthorized mass use of works by composers, lyricists, and performers.

Their stance is that if AI companies develop commercial services using protected works, they need to disclose what content was used, respect rights, and return profits to creators.

This is a perspective that does not pit technological innovation against rights protection.

Rather than halting AI development, the idea is to establish licensed training data and distribute the revenue from AI services to rights holders. This approach aims to extend the rights processing mechanisms built in music distribution services, broadcasting, and video platforms to AI learning and generation.

Voices on Social Media: "An Expected Judgment" and "Unclear Boundaries"

Immediately after the ruling, expressions like "GEMA won a complete victory over Suno" and "Legal brakes on unauthorized AI learning" spread in the Japanese-speaking X.

However, the court's official announcement stated that "GEMA's claims were largely upheld," and the expression "complete victory" on social media is somewhat simplified. Also, the ruling is at the first instance and not yet finalized. Suno has also expressed disagreement with the decision and is considering options, including an appeal.

Notable in public posts was the welcoming reaction as a decision to protect creators' rights.

The opinion is that if AI services can learn popular songs without permission and create many similar outputs in a short time, it is unfair not to pay those who created the works. Users believed to be composers also posted expectations for a stronger trend towards seeking permission for AI learning.

On the other hand, while understanding the direction of the ruling, some questioned how much reproduction is needed to determine that "there is a reproduction within the model."

For example, the evaluation might differ between cases where similar music emerges only after numerous special instructions and cases where something close to the original appears with a simple single instruction. The practical delineation of similarity judgment technology, the number of trials needed for reproduction, and the specificity of input content are not yet fully visible.

Furthermore, there was a cautious reaction that "without reading the full judgment, it cannot be applied to all general AI learning."

These reactions indicate that even on social media, the discussion is not simply a binary opposition of pro- or anti-AI, but rather a more specific debate on "how to measure reproducibility" and "how to distinguish between learning general styles and memorizing specific works."

Is AI Learning Widely Accepted in Japan?

The main reason this ruling is attracting attention in Japan is that the legal system surrounding AI learning and copyright is not the same as in Europe.

Article 30-4 of Japan's Copyright Act stipulates that copyrighted works can be used within a certain scope for information analysis not intended for enjoying the thoughts or emotions expressed in the works. AI learning has been considered potentially subject to this provision.

Therefore, it is sometimes explained that "AI learning is basically free in Japan."

However, in reality, not everything is unconditionally allowed. If the purpose of enjoying the work coexists with learning or if it unjustly harms the interests of the rights holder, the limitation provisions may not apply.

Moreover, even if the use during the learning phase is permitted, the generation and use phases are judged separately.

If AI generates music similar to existing works and is recognized as relying on those works, it could be problematic, similar to ordinary copyright infringement. When companies use AI-generated music for advertisements, games, videos, or store BGM, it is dangerous to assume "it's safe because AI made it."

The Agency for Cultural Affairs also publishes checklists and guidance for AI developers, service providers, users, and rights holders, seeking responses to reduce risks.

The German Ruling Aligns with JASRAC's Claims

The Japanese Society for Rights of Authors, Composers and Publishers (JASRAC) has long argued for the need to protect human creativity and the "cycle of creation" in relation to generative AI and creation.

If creators' works are used for AI learning on a scale and speed incomparable to humans, and the resulting generative products replace existing works, the economic foundation for creating new works could weaken. If unauthorized use by commercial companies becomes the norm, it could lead to free-riding on creators' efforts.

The recent German ruling aligns with many of these concerns raised by JASRAC.

In particular, the issue of whether it is appropriate to treat it as mere information analysis when works are reproducible within the AI model and the actual output competes with the market of the original songs could influence the interpretation of Japan's Article 30-4.

Of course, a German first-instance ruling does not directly bind Japanese courts. However, AI services are provided across borders. Operating with licensed data for Europe and unlicensed data for Japan is socially and commercially difficult to maintain.

Companies providing services in the international market will likely be required to manage data and licensing systems that can be explained across multiple regions, not just based on the country with the least regulation.

Is This a Tailwind for Japanese Musicians?

From the creators' perspective, this ruling could be a tailwind for advancing rights negotiations.

Until now, it has been difficult for individual composers and lyricists to verify whether their works were used for AI learning against large AI companies. If the training data is not disclosed, gathering evidence of rights infringement is also not easy.

A system where copyright management organizations negotiate on behalf of their members, litigate if necessary, and demand disclosure of usage information and revenue from AI companies could reduce the burden on individual creators.

At the same time, it is not enough for rights holders to simply reject AI use uniformly.

Organizing licensed songs, sound sources, and metadata and providing datasets that AI development companies can legally use will become important. GEMA has also started providing rights-processed AI training music datasets.

If a market is established where AI companies purchase transparent data and pay compensation according to usage and revenue, AI could become not only an enemy but also a new source of income for creators.

AI Using Companies Will Also Need "Source Verification"

The impact is not limited to AI development companies and musicians.

It concerns all businesses that use AI-generated music, including advertising agencies, video production companies, game companies, broadcasters, store operators, and influencers.

Companies should verify not only the commercial use permissions in the terms of use but also whether the service provider itself infringes on third-party rights.

In the future, it will be necessary to confirm whether the AI service being used has licensed training data, whether there is a system for rights holders to request deletion, whether similarity to existing songs is checked, and whether compensation or response is available in case of problems.

Those publishing AI-generated music will also need to avoid using specific artist names or song titles in instructions, subject the completed audio to similarity checks, and record the production process and input content.

The Question Is Not AI's Existence but the Distribution of Profits

In discussions about generative AI, arguments like "don't stop technology" and "protect human works" often clash.

However, what emerges from the ruling and the SPD's reaction is a more realistic third way.

Rather than banning AI music itself, the idea is to make training data transparent, obtain permission from rights holders, and distribute the economic value created by AI to creators. The issue is not AI creating music but the exclusion of those who provided the foundation from the value chain.

Human musicians also listen to past works and create new music under their influence. However, the scale, speed, and economic power of human learning compared to companies mechanically reproducing millions of songs and building global commercial services are different.

Ignoring this difference and treating AI learning as entirely the same as human creative activity is unreasonable. On the other hand, a method to objectively measure how much AI models remember works is still in development.

This ruling does not declare all AI learning illegal. It focused on the circumstances where specific works were held in a reproducible form within the model, and outputs essentially similar to the original were obtained from simple instructions.

Therefore, the ruling should not be read as "a complete ban on AI." At the same time, the simple explanation of "it's free because it's learning" will no longer be easily accepted.

What Japan needs is not to choose between regulation or freedom but to specify what kind of learning requires permission, what kind of output constitutes rights infringement, and how to distribute profits.

The German decision regarding Suno has advanced this discussion to the next level.

As long as AI learns from human creations, respect for musicians must be demonstrated not only in principle but also through mechanisms of data transparency, licensing agreements, and compensation payments. Whether generative AI can become an entity that enriches music culture depends not only on technological performance but also on the ability to design a fair market.

Source URL

1. An article reporting that the German Social Democratic Party welcomed the Suno ruling and called for transparency, respect for rights, and fair compensation for creators
https