Rachel’s Song – Part I: Rights of authors to music created by or with artificial intelligence under German copyright law


In one of the many iconic scenes of Ridley Scott’s movie Blade Runner (1982), an artificial “replicant” called Rachel plays piano for the detective that is supposed to hunt her down and “retire” her, making him question his own humanity. More than 40 years later, music created by or with artificial intelligence (“AI”) is omnipresent. Both creators and users often assume that works generated by AI cannot enjoy copyright protection and, therefore, may be used freely. When it comes to music, this is not the necessarily true. While an AI system cannot be “author” under German copyright law, creators of music can gain exclusive rights to compositions and recordings generated by or with AI technology. Users should carefully check who was involved in the specific creative process and to what extent – and protect themselves with proper licence agreements.

This Part I of the article examines under what circumstances those involved in the creation of music by or with the help of AI programs can acquire rights as authors. As a rule, there are no authors’ rights in music that is created autonomously by generative AI systems. However, various applications of AI technology leave music creators sufficient creative leeway to obtain copyrights in the resulting works. Part II focuses on potential neighbouring rights in recordings, performances, broadcasts or databases involving music created by or with AI programmes.

Only humans can be authors

Machines and computers cannot be “author” within the meaning of German copyright law. Sec. 7 of the German Copyright Act (Urheberrechtsgesetz; “UrhG”) enshrines the creator principle (Schöpferprinzip) according to which the making of protected works requires an act of creation. A system of artificial intelligence (“AI) – no matter how autonomous and highly developed it is – cannot perform such act. This is because it requires a factual act independent of intellectual capacities that only humans (including children and mentally disabled persons) are capable of: The creative process must result in the individual spirit and originality of the creator being reflected in the work.[1] Only human creators can have this kind of personal-spiritual connection to their works. The law protects the non-economic and reputational interests by granting the human author moral rights.

Authors are typically protected when using AI technology as a mere tool

Music creators nevertheless can obtain rights when using AI technology. This is the case to the extent authors use any existing creative leeway to express their personality and originality.[2] In practice, creators of music have been employing a wide variety of AI programmes for many years – some as a source of inspiration and some as digital tools within the creative process. This does not jeopardize their authorship. A recent example is the Beatles song “Now and Then”, which was released in last November. A vocal track of the late John Lennon was extracted from an old demo recording and implemented into the modern production by means of AI technology. This technique has no effect on the copyrights of John Lennon’s heirs or the other participants. Only the human contributions shape the individual character of the song.

Autonomously AI-generated contributions are generally not protected – but there are exceptions

The situation is different where (parts of) musical works or audio tracks are created entirely by systems of generative AI (“GenAI”). In these cases, there is typically no room for the persons involved to express their creative individuality as these systems autonomously make use of the existing creative leeway:

  • Neither the development nor the training of common GenAI programmes enable sufficient room to creatively influence the individual output.[3]
  • The mere activating and operating of machines or programmes that autonomously produce works is not sufficient to establish authorship.[4]
  • The same applies – in general – to the input of the AI user’s individual instructions. These so-called prompts are usually not detailed and/or determined enough to control the copyright relevant elements of the output. This especially applies to AI programmes that convert mere text input into music, such as Suno or Loudly. Most GenAI systems are designed and trained in such a way that, even when receiving detailed specifications, they still have an almost infinite number of possibilities to shape the individual output. This may only be different for models that have been trained with a relatively small amount of data.
  • Finally, the mere selecting of individual pieces from a greater number of AI-generated content does not justify copyright protection.

In some cases, the output of GenAI systems can still contain copyright-protected content. Services such as Google’s Tone Transfer or Mubert Studio enable users to upload audio tracks or samples as prompts. These may include content belonging to the person entering the prompt or to a third party. If (relevant parts of) the pre-existing works continue to exist in the output, its use must be authorized by the authors of such works. Whether this is the case must be examined on a case-by-case basis in accordance with the CJEU’s criterion of recognisability.[5] For example, if an AI programme is provided with a vocal track including the main melody, the creative originality of the author of this melody is likely to be reflected in most of the versions of the generated output – irrespective of any autonomously added tracks or elements.

Furthermore, autonomously AI generated sound sequences can be used by humans to create a new independent musical work by means of sampling. It is irrelevant whether the used samples are protected. The creative achievement consists of the “artistic use of sampling as a tool of musical creation”.[6] In this way, hip hop and electronic music artists may avoid burdensome rights clearances and licence fees while maintaining their own rights in the music.

Finally, collection works and database works under Sec. 4 UrhG can be compiled from (parts of) AI-generated music. Such works are protected if the selection and/or arrangement of the individual elements of the collection passes the threshold of a personal intellectual creation.[7] The individual elements, on the other hand, do not have to qualify as a work. However, it should be noted that the non-protected elements of such a collection or database can still be freely used – separately as well as in other combinations.

Human contributions to AI-generated works can be protected

If certain parts of a musical work are created by a human and others are created independently by an AI, copyright protection requires that the individual human contribution qualifies as a personal intellectual creation. Humans cannot make the AI-generated elements their own.

The threshold for protection of musical works and associated lyrical works is rather low:

  • The originality of music is primarily based on melodies. In addition, authors can express their individual creativity by using the “expressive means of rhythm, tempo, harmony and arrangement”, as well as “instrumentation and orchestration”.[8] In contrast, it is not sufficient if the human contribution is limited to the use of purely technical and/or common musical means, such as simple sequences of tones or well-known rhythmic structures.[9] A fortiori, individual notes, chords or the sound of an instrument, cannot enjoy independent protection.
  • Popular music typically consists of a combination of musical works and song lyrics. The individuality of linguistic works can result from their content or from their form including style and arrangement.[10] Human-created lyrics regularly enjoy protection. However, banal texts not going beyond clichéd metaphors and the usual forms of expression can fail to qualify for protection.[11] The shorter the text, the more likely it is to lack the required level of creativity. Music and lyrics of a song are combined works that can be exploited together or separately. If one of both is generated autonomously by AI, the author of the other has no control over the separate exploitation of the non-protected AI contribution by third parties.

The legal assessment may become more complex where the contributions of humans and GenAI to a musical work are so intertwined that they cannot be separated from each other and exploited separately. As with works that are created jointly by several people, only persons whose contribution in itself qualifies as a personal intellectual creation can acquire rights to the overall work. The standards established by the courts on co-authorship can be applied to assess the protectability of a human contribution to AI music. In exceptional cases, the human contribution may be found to have such a prevailing influence on the character of the work that the entire creation can be attributed to him/her. However, rather subordinate or mere technical contributions cannot result in AI-generated parts being legally attributed to the human contributor. Until the courts have provided further clarity here, the licensing parties, in such scenarios, should pay particular attention to the contractual protection of their interests.

Finally, humans can also acquire author rights through the (subsequent) adaption or arrangement of AI music. This requires that the alteration per se qualifies as a personal intellectual creation. The adaption must create an overall musical and aesthetic impression that was not present in the original work. The Federal Supreme Court recognises that changes to the overall structure, the melody, the tempo and the arrangement can be sufficient expressions of the adapter’s creativity.[12] The same applies to modifications of the instrumentation, orchestration, or rhythmization.[13] However, minor changes in melody, rhythm or harmony merely made to optimize the playability without affecting the character of the original music, will usually not be sufficient to gain protection. It must also be noted that, irrespective of the existence of a copyright protectable adaption, the non-protected AI-generated original can still be used freely by any third party.

Conclusions from a practical perspective

If creators of musical works use AI technology as a source of inspiration or as a mere tool within the complex creative process, there is no need for any special legal precautions. However, creators and users of music should both be aware of the particularities of copyright law when autonomous GenAI systems are involved. In those cases, appropriate documentation and contractual arrangements are key for safeguarding their mutual interests:

  • The more fragmented the individual steps in the creative process are and the more human and artificial elements are blended, the more important it becomes for all parties involved that creative contributions are documented in detail.
  • Where copyright protected works are used as input, substantial parts may still be recognisable in the AI generated output resulting in a need to license such pre-existing works before usage.
  • In cases where human and AI parts are intertwined or AI generated music is sampled, adapted, or rearranged by humans, copyright protection depends on whether the human contribution per se passes the threshold of sufficient personal creativity. Generally, the principles already developed by non-AI related case law can be applied here. Remaining uncertainties should be addressed in licensing contracts.

[1] CJEU Case C‑145/10, para. 88 et seq. – Painer.

[2] CJEU Case C-683/17, para. 31/32 – Cofemel.

[3] The training of GenAI with pre-existing works may infringe third-party copyrights.

[4] Regional Court Berlin GRUR 1990, 270 – Satellitenfoto.

[5] CJEU Case C-476/17, para. 31 et seq. – Pelham; Federal Supreme Court GRUR 2022, 899, 905 – Porsche 911.

[6] Federal Constitutional Court GRUR 2016, 690, 693 – Metall auf Metall.

[7] CJEU Case C‑604/10, para. 29 – Football Dataco.

[8] Federal Supreme Court GRUR 2015, 1189, 1192 – Goldrapper.

[9] Federal Supreme Court GRUR 1988, 810, 811 – Fantasy.

[10] Federal Supreme Court GRUR 2011, 134, 137 – Perlentaucher.

[11] Federal Supreme Court GRUR 1991, 531, 532 – Brown Girl I.

[12] Federal Supreme Court GRUR 1991, 533, 534 – Brown Girl II.

[13] Federal Supreme Court GRUR 1968, 321, 324 – Haselnuß.

Leave a Reply