How AI Rights Are Changing Record Contracts

This March, Jimmy Hession, an artist manager at Milk & Honey, was ironing out a producer agreement with the Sony-owned, European dance label B1 Recordings for his client Paul Harris, who worked on the Alok and Khalid track “Dive into Me.” When Sony Music Germany sent over the deal terms, Hession discovered some novel language: a broad release allowing the use of Harris’ work for AI training.  

Specifically, the contract granted the label the “unlimited, exclusive rights” for the company to “use the recording in models and systems of generative artificial intelligence and applications based thereon, including generative AI, including [but] not limited to the analysis of the Recording for the purpose of extracting information on patterns, trends and correlations (AI training).”  

Hession pushed back on this provision, and the two parties ultimately reached a compromise, giving Hession’s client some approval rights for the use of his recording in AI training. Crucially, though, these approval rights do not extend to any “blanket license” that Sony may sign to grant another party access to “all or a significant portion” of its catalog.  

Hession’s experience was not an isolated incident. Instead, sources tell Billboard it is part of a growing new trend in music dealmaking where labels are trying to accumulate AI training rights — either through explicit new contract language or novel applications of long-standing licensing provisions.

B1 Recordings, for example, included near-identical wording in a different artist single agreement this past November, according to a U.S.-based artist attorney. French music giant Believe put a provision in an October 2025 distribution contract which allows it to license their content in datasets to “research, train, develop [and] test gen AI models and/or products.” In a BMG distribution deal, obtained by Billboard via a European music executive and translated from German, the contract phrasing mentions AI in multiple contexts. This includes an “AI Right,” which refers to the right to use songs created during the contract term in whole or in part in connection with artificial intelligence systems, “in particular to feed the contractual products into an AI system as training, validation or test datasets” and to “exploit the output generated by the AI system.” 

That last contract also forbids the artist to turn in AI-manipulated or AI-generated tracks as part of the delivery requirement for the BMG deal; prohibits the creation of re-recordings (like Taylor Swift’s), including re-records that use AI to deepfake or otherwise simulate the artist’s voice; and grants BMG the right of adaptation — meaning remixes, sampling or other modifications, including those that use AI — and to exploit those adaptations.  

The addition of these novel contractual provisions around AI training and AI use in general appear to have coincided with the music industry’s move towards partnering with the new wave of AI music startups in recent months and settling parts of their lawsuits against them. In the fall, Universal Music Group (UMG) and Warner Music Group (WMG) both settled their claims in the blockbuster copyright infringement lawsuit against AI music company Udio and entered into licensing deals to “[create] new revenue streams for artists and songwriters, while ensuring their work remains protected,” as a press release about Warner’s deal stated. Kobalt and Merlin followed suit with Udio deals months later.

In December, WMG followed up with a settlement and licensing agreement with Suno, which had also been hit with a near-identical copyright infringement lawsuit to Udio’s. (Sony Music, the other major music company who sued Suno and Udio along with UMG and WMG, has not yet announced any deals with the two firms and remains in active litigation.) 

In these settlement announcements, music companies have made a point to state that these deals ensure artists’ work is licensed and authorized. The natural result is that AI training has now become an element of record deal negotiations — and sometimes a significant pain point.

At times, this comes in the form of labels and distributors drafting clear language that grants them the right to use content for AI training, as in the case of the European deals identified by Billboard. Talent lawyer Avi Dahan says he’s been presented with this kind of provision in some recent contracts, though there “isn’t necessarily an industry norm yet around these points.” 

Dahan describes the current dealmaking landscape as the “Wild West” — a transition period where it’s becoming common practice for labels to require the disclosure of AI used to create a recording, but only some are asserting rights to use music for AI training of their own.  

Colin Morrissey, an artist lawyer at Granderson Des Rochers, similarly says AI training clauses are “starting to creep their way into some new recording agreements” from smaller distributors and technologically-driven music companies. According to Morrissey, though, it’s still “pretty rare” to see an explicit grant of AI training rights in major-label deals.  

It’s likely not a coincidence that all the examples of explicit AI training language identified by Billboard were in German and French recording contracts. According to Estelle Derclay, a professor of intellectual property law at the University of Nottingham, many European countries have laws requiring contracting parties “to spell out exactly all of the rights that you give, not just in a general clause.”

This is different from U.S. and U.K. law, which allow for broad contract provisions in which artists often grant their record labels a nonspecific right to exploit their songs. It’s common for such record deals to include clauses granting the right to use an artist’s work in “blanket licenses” of a label’s entire catalog, such as deals with social media companies like TikTok and Instagram.

Multiple top artist attorneys tell Billboard they’ve recently begun to realize that labels could feasibly use “blanket license” clauses in U.S. record deals to opt-in artists’ works to train their AI partners’ models without seeking individual artist approval. 

Often, when deals between AI companies and music companies are being discussed, executives for both sides will reference what’s known as “opt-in,” a term used to describe the process of allowing artists to choose whether or not they participate in AI licensing. (That stands as opposed to “opt-out,” meaning that artists are automatically included in AI licensing unless they elect to take themselves out of it.) Over time, this has become an overwhelmingly popular stance in the music industry to assure that creatives will maintain their agency in an age of AI.

WMG CEO Robert Kyncl even called “opt-in” one of his three non-negotiable principles in a blog post in November 2025, saying that “artists and songwriters will have a choice to opt-in to any use of their name, image, likeness, or voice in new AI-generated songs.” Udio CEO Andrew Sanchez also referenced opt-ins in reply to a question from Billboard about his licensing deal with UMG, saying “[Udio is] not just [going to offer] remixing and mashing up. It’s also creating in the style of artists with their opt-in.” UMG’s chief digital officer Michael Nash also brought up opt-ins on Billboard’s On the Record podcast, saying, “We didn’t think that made any sense [to work with AI companies, like Suno, who let AI songs leave their platforms], [and] we didn’t feel like when going to artists for their support to opt-in to these services that it was going to be a very compelling pitch.”

Sony declined to comment for this story. BMG, UMG, Merlin, Kobalt and WMG did not respond to requests for comment. A representative for Believe responded with the following statement: 

“Believe and TuneCore’s GenAI policy is governed by the core principles of Consent, Control, Compensation, and Transparency, ensuring that generative AI remains a tool for artist empowerment rather than unauthorized exploitation. Central to this approach is a voluntary opt-in mechanism for the GenAI initiatives we propose to our artists and labels, which provides rights protection and early access to ethical AI opportunities. Contrary to concerns regarding the use of blanket licenses for training, Believe requires explicit opt-in to the programme and explicit consent on a case-by-case basis before the formal launch of projects involving artist content, and applies a defined scope, safeguards and transparency framework to any use of content, providing the basis for an informed decision.”

While many believe the music companies’ interest in using the “opt-in” approach was to allow artists the option both to control if their work is used in AI training — or inputs — and AI-generated outputs, such announcements seemingly never reference training specifically. And that has led to some suspicion among artist advocates.

“We’re seeing a differentiation between the way training — or inputs — and outputs are treated,” says Audrey Benoualid, partner at Myman Greenspan Fox Rosenberg Mobasser Younger & Light, pointing to the potential use of blanket licensing for AI training. Jason Boyarski, founding partner at Boyarski Fritz, adds, “Some of the labels have already taken the position that they technically don’t need special approvals to train.”

“As of now, the grant of rights in deals are broad enough to allow them to do it,” says Morrissey of AI training. In response, Morrissey says he’s “focused on the approval rights for our artist clients around not allowing their music to be trained in AI tools. We kind of assume, a lot of time, that the label will have rights in these sorts of cases and try to work back from there to give our client as much approval over that [as possible].” Benoualid says she has also had success in asking for approval rights, “depending on the artist.”

Derclay notes that artists in the U.S. and U.K. might have the right already to nix AI training uses of their work if their record deals include approval rights for “future technology.” Otherwise, they can try to win approval rights now through contract re-negotiations. 

Still, just because music companies could potentially rely on a blanket license to circumvent the need for individual artist approval for AI training, that doesn’t necessarily mean they will. Morrissey says he thinks “a  lot of these labels are going to ask regardless [of approval rights] because it is such a hot-button issue.” Boyarski holds a similar view, citing the importance of maintaining a positive artist-label relationship. “I don’t think they will [opt artists in without approval],” Boyarski says. “I think the labels genuinely want buy-in from the artist community in a way that makes it work. I think that there’s going to be a partnership between the artists and labels to find the right solution.”