Who owns an A.I.-created song?

Normally, writing songs is a bit like building a sandcastle. In this scenario, you start with nothing but just a flat surface at the beach or in a playground sandbox, and build up from there. While you might use a bucket, shovel or any number of different tools to help you craft the castle, it’s ultimately all of your own doing. Similarly, while a songwriter might come to the writing session with years worth of experience and creative inspiration, and might use a variety of physical and digital instruments and tools to facilitate their writing process, they usually start with a blank sheet of paper or a brand-new DAW session, fighting to build up a new song from nothing.

With the assistance of artificial intelligence, however, writing a song might be more akin to carving a sculpture out of stone or marble. Just as nature produces a billion different rocks on its own without human intervention, a compositional algorithm can generate a seemingly infinite number of melodies, chord progression or beats with the push of a button. An artist then plays the equal role of curator and creator, picking the best musical “rock” to start chipping away at, eventually carving the amorphous object into something recognizable as the artist’s own, distinct creation.

In the sandcastle example, it’s clear that the equipment used to write and record the song does not own the song itself. Such equipment was merely a vessel for expressing the songwriter’s ideas and vision; assuming no sampling or plagiarism was involved, the songwriter(s) did all of that visionary work, so they own the totality of the song’s copyright.

On the other hand, if a developer is creating a program to produce songs automatically based on preexisting, ingested training material, that a second person is then altering further into a different creation, it’s much less clear who should own that final song.

Because the ecosystem is so relatively new, there’s no legal standard for whom exactly to credit in a songwriting and production process that incorporates A.I. For instance, in Warner Music Group’s controversial deal with generative audio startup Endel, the app’s software engineers are credited as co-writers alongside the composers who made the original source material. In contrast, the creative apps built through Google Magenta — a research project that uses Google’s machine-learning platform TensorFlow to make art — are open-source and framed merely as tools, not as coauthors. Put another way, there is no expectation for artists to credit or turn over creative ownership to Google when they release music made with Magenta’s tools.

So to answer the question posed in this headline — “Who owns an AI-created song by default?” — there is no such default, at least not yet. That said, startups in the emerging field of automated music creation software are setting important precedents in their stances on ownership that may fundamentally change how artists relate to the technology in the long term.

The case of Boomy

The first step in figuring out who will own the song created with any A.I. tool is to read the company’s end user license agreement (EULA), which will likely have a dedicated section for intellectual property.

Let’s start with Boomy as an example. Boomy is a music creation tool powered by artificial intelligence, allowing users to produce new musical material with the push of a button. While there is little that the users themselves are composing from scratch, they can still modify different compositional elements (e.g. lead/bass/drum density, tempo) on a series of sliders, change the instrumentation and mixing style of the output (e.g. “New York Garage” vs. “Vinyl Lofi”) and even add their own vocals, which Boomy’s A.I. will transform as it sees fit to match the underlying instrumental. Part of the mobile workflow for Boomy is displayed in the screenshot at the top of this article.

Unlike most other creative A.I. music tools, Boomy also allows users to distribute and monetize their creations on streaming platforms like Spotify and Apple Music and social networks like TikTok, Facebook and Instagram for an additional fee ($2.99/month for up to 20 albums or singles, and $9.99/month for unlimited albums and singles).

Who owns a song made on Boomy? The answer might surprise you.

If an app like Boomy is just a creative tool that invites user manipulation, and ultimately full user discretion over which songs are “good,” it could be argued that the user pushing the buttons on the site and setting the parameters of the songs ultimately owns the underlying copyright. But this is not the case with Boomy: Under their end user license agreement, the ownership of all songs created on the platform “is retained by Boomy, but the user is granted a license to use that song for several purposes.” The user can download their own Boomy song in MP3 form and monetize it on Spotify and Apple Music under a paid plan, but they do not actually own any part of that song.

“The way we work is that rights are assigned to Boomy by default and then depending on what the user needs, we either manage those rights on their behalf, sell them 100% of the ownership rights for a small fee, or work out something in between if they have a complex use case — i.e. label or publisher agreements to contend with,” Alex Mitchell, founder and CEO of Boomy, tells me. “You make songs with our tech for free, then we own, manage and distribute them and pay you up to 80%* of what they make. If you need the ownership rights for some reason, that’s cool too, just let us know and we’ll send you the rights agreement.”

In most cases, Boomy will let users buy out the copyright ownership of their songs for a flat fee of $19, with exact rights and revenue splits determined on a case-by-case basis. “The agreement may require specific language depending on the use case and territory,” says Mitchell. “It is burdensome to do these agreements on an individual basis, but we do it because certain use cases need certain rights and we’re happy to support those use cases.”

While Boomy’s EULA makes clear that they, not their users, own all the music created on their platform from the outset, it’s not immediately clear why this is the case. After all, the equivalent stance would arguably be a digital audio workstation (DAW) like GarageBand or Ableton taking full ownership of all the music created on their apps, which would be absurd.

Mitchell is upfront about Boomy’s intentions: He argues that this ownership path not only helps the company avoid messy copyright entanglements, but also provides the company with additional revenue in a way that doesn’t inherently hinder the free user experience, unlike other forms of revenue such as advertising. Assuming default ownership can also help Boomy maintain a stricter level of moderation and keep its technology from being used for hate speech or other unsavory content, which it can request to be taken down at any point.

“To be clear, my eventual goal is to have policy and law catch up to us so that we can clearly designate a cut of the masters & publishing ownership rights to users” on a global scale, says Mitchell. “In my mind, giving a user a cut of the ownership rights could be a more preferable model to what we’re doing today, so long as we could still ensure ethical use of the technology.”

Platforms setting precedents

By “policy and law,” Mitchell is referring above to how there is also no standard for A.I. copyright ownership at the level of national governments. For instance, computer-generated works can benefit from copyright protection in the U.K., but there is no such clarification in the U.S. Thus all agreements that companies like Boomy make in transferring rights to users have to be on an individual basis, and take time and money to execute.

Boomy’s EULA demonstrates a compromise that software startups in this sector might need to make in order to remain operational within these existing legal constraints. In fact, Boomy is not the only company claiming full ownership over its creations. Popgun Labs, which develops the A.I. music creation software Splash, gives users a royalty-free, all-purpose (i.e. commercial and non-commercial) license to “use the sound recordings you create using the Software.” But users still do not own any of the instrumentals created with the software, and Splash’s terms of service dictate that “your use of the Software and the Content does not transfer to you ownership of any IP Rights in the Software or the Content.”

When asked if users will be confused by why they may not own the songs they created with help from artificial intelligence, Popgun Labs founder Stephen Philips tells me bluntly: “Yes. It is super confusing for everyone already.”

Strategically, though, this stance makes sense for companies even if it might not be the most end user-friendly. Companies that invest in, develop and maintain creative A.I. software over time “will probably want to claim ownership over anything that system generates, including any transcriptions and derivative works,” Gregor Pryor, partner and co-chair of the Entertainment and Media Industry Group at Reed Smith, tells me. “A.I. software generates content and companies will want to create value [around] and exploit that content as an owner, most likely granting limited licenses to users.” Startups like the aforementioned Endel already exploit machine-generated music and audio in a way that keeps ownership exclusive to the organization.

A major legal aspect of whether a platform should be claiming 100% ownership over songs created with its A.I. tools boils down to the forever contentious debate around “originality” in the creative process. “One factor to consider is the extent to which the A.I.-created song really was created by the A.I. engine, or whether human intervention was critical,” says Pryor. “For music to gain copyright protection, it has to be original. To assess if it is original, you need to look at whether and how the algorithms and datasets for training the A.I. system use copyrighted works. Some A.I. software just generates raw material that needs significant editing and producing before you reach the final product.”

The notion of originality marks a crucial difference between Boomy and some other projects like OpenAI’s Jukebox, which allows users to generate music, lyrics and vocals in the styles not only of certain genres like country, pop and heavy metal, but also of specific artists like Elvis Presley, Katy Perry and Frank Sinatra.

Jukebox clearly uses existing, copyrighted work to train and directly influence its (very technically impressive) creative output, which would make it hard for OpenAI to claim 100% ownership over its songs. As Jukebox’s license on GitHub states: “We don’t claim ownership of the content you create with Jukebox. We only ask that you use Jukebox responsibly and clearly indicate your content was created using OpenAI’s Jukebox.” In addition, unlike with Boomy, OpenAI does not give users a commercial license to monetize songs made with Jukebox: “No portion of the Software, nor any content created with the Software, may be used for commercial purposes.”

In contrast, Boomy claims to be generating 100% “original” content — meaning it is not using any preexisting copyrighted content in the training material for its systems, and only uses Boomy-created songs to train the system’s future outputs.

“I am compelled to clarify that Boomy is not an ‘A.I. song creator’ in the same way that OpenAI or Magenta are,” says Mitchell. “It’s an annoying but important distinction. At Boomy, people make music with A.I.-powered tech … All 650,000-plus tracks created so far are human work that’s been assisted by automation. Boomy users aren’t making ‘A.I. music,’ which generally refers to systems that accept training data and regurgitate best-guess copies, like OpenAI.”

Everyone, not just musicians

Legal considerations aside, a major promise behind A.I.-powered music creation software is broad accessibility. Similar to how the internet opened up music distribution on a global scale, these new kinds of tools can democratize music creation. Mitchell tells me that 75% of Boomy users had never made a song before using the tool, and that 20% of its users use Boomy as their only outlet for making music.

As these tools continue to rise, the definition of who is or isn’t an “artist” or “musician” is going to change rapidly. What won’t change is how confusing and ambiguous copyright law remains in the U.S. and around the world. Thanks to technology, the market is only further expanding for who needs to understand these laws — whether it’s musicians who are highly skilled at a specific instrument, or everyday people who make music with more automated software.

Upon seeing Boomy’s EULA, I was initially startled. However, after spending enough time considering it and hearing why it’s in place, it seems reasonable. A.I. music creation platforms today seem to be making choices on two extremes: Immediately grant users all the rights, or immediately assign those rights instead to the software company. If the rights go to the user first, then the service will (likely) need to find other ways to monetize its technology. If the service is free and open, it runs the risk of being used in unintended and malicious ways. Having the business assume ownership in a more open scenario seems fair and reasonable — at least to me.

In any case, the most important takeaway for any company in this sector is the value of clear communication. When Instagram changed its terms of use in 2012 to state that it could use people’s photos without their consent, it understandably created a controversy. No one likes feeling like the rug is being pulled out from underneath them. In Instagram’s case, the problem could have been avoided with better messaging.

A.I. music tools will need to be prepared for this public reckoning, too. It would not be a good look to take legal action against a user who didn’t understand where and why a song they made could be used. Whatever path a company follows in terms of copyright ownership, they need to make sure it’s communicated very clearly to the user, and encourages rather than stifles a new generation of creativity.