AI-generated Drake and The Weeknd song removed from Spotify and Apple - Can UK copyright be enforced against AI-generated songs?

The song, Heart On My Sleeve, features cloned voices of Drake and The Weeknd. The creator, known as @ghostwriter, has claimed the song was created by AI that was trained on the artists’ voices.

The song went viral since its release last Friday (14 April 2023) – from then until Tuesday of this week, the song was listened to at least 8.5 million times on TikTok. It has since been pulled from Apple Music, Spotify, Deezer and Tidal. It is also being pulled from TikTok and YouTube, but there are still versions of it available on these platforms. The removals have been on copyright grounds.

From a UK perspective, the creative industries have concerns about the use of AI. For example, the recently proposed introduction of a permissive, AI-friendly, text and data mining exception was abandoned following resistance from those industries and parliamentary debate.

One of the many objections to the proposal is almost identical to the issue here: broadening the exception would allow AI to create songs at a rapid rate based on an artist’s own work. Essentially, the artist would be competing with a faster version of himself/herself. As a result, artists would be dissuaded from creating original content. 

The minsters’ debate on the UK’s, now abandoned, text and data mining exception to copyright seemed to be grounded on the implicit assumption that the current law was sufficient to address these challenges. So what is the position? Does an AI-generated song infringe copyright in the UK?

In this example, if indeed the AI song is only replicating the voices of Drake and The Weeknd to create its own original work, it could be difficult to demonstrate that the AI song copies one or more of the original songs of the respective artists. Indeed, they would perhaps need to rely on performance rights, which would come with its own hurdles. In particular, it may be difficult to argue that a specific performance is being copied.

Moving away from this specific example to a more general example of an AI-generated song copied from one or more original songs, there are two main potential infringing acts involved in the creation of an AI song that should be separated.

The first relates to the input data: would the initial copying of the original song(s) onto servers to train the AI amount to a copyright infringing act? This is more straightforward from an infringement perspective – since the whole of the original song would be copied onto a server. However, it may be difficult to show that UK jurisdiction would bite. For example, if the original song is stored on servers that are based abroad, and the copying entity’s servers are also based abroad, it may be difficult to argue that the UK courts have jurisdiction.

The second concerns the output data (the AI song): would the AI song itself infringe an original song, or several original songs, of the artist? If the AI song was made available in (and targeted to) the UK, the UK courts should have the necessary jurisdiction. However, from an infringement perspective, the copyright owner would need to demonstrate that a substantial part, that is to say an important part, of their original song has been reproduced in the AI song. Depending on the circumstances, this could be hard to demonstrate, as the AI will have taken many songs into account from many artists.

There are of course many other questions and legal issues involved in such scenarios. For example, as AI becomes more autonomous, who is the correct entity to bring an action against where human involvement is limited or non-existent? Will moral rights be engaged where AI mimics a performance of an artist?

In response to Sir Patrick Vallance’s recent Pro-Innovation Regulation of Technologies Review (see here), the UK Intellectual Property Office has signalled its intention to issue a Code of Practice in the summer to provide guidance to AI developers on accessing, and using, copyright-protected works as inputs to train their models, and also ensuring “protections” on generated output to “support right holders of copyrighted work”. The response goes on to state that an AI firm committing to the code can “expect to be able to have a reasonable licence offered by a rights holder in return”. Watch this space.