Spot the “Covenant Not to Sue” trap in contracts

The AI “License” Trap: A Musician’s Guide to Navigating the Legal Minefield of Generative Music Deals

As generative artificial intelligence models chase financing and growth, music companies are entering the negotiation table with licensing agreements—and the artists who provide the raw material. On the surface, these deals sound promising, suggesting a fair exchange for creative work. But beneath the glossy presentation of “licensing,” a complex legal trap is often hidden. For musicians, this isn’t just about royalties; it’s about surrendering rights that can be far more valuable than the money on the contract.

The central tension lies in the fine print. Many agreements are not true licenses but rather elaborate pieces of litigation insurance dressed up in legal jargon. The key to surviving this negotiation is understanding the difference between granting permission and permanently surrendering your ability to enforce your rights. As the old adage suggests, the large print gives you one story, while the small print holds the truth.

One of the most critical elements to scrutinize is the concept of a covenant not to sue. This is a contractual promise that you will never bring a legal claim against the other party, regardless of future circumstances or whether you even know you have a claim. Unlike a standard license, which grants permission and can expire, a covenant not to sue functions as a permanent waiver of your right to seek remedies. Signing this means potentially giving up rights over copyright, publicity, moral rights, and potential future lawsuits—effectively surrendering control forever.

Another pitfall involves co-writers and shared ownership. If an agreement requires you to promise that no third party needs to consent, it can inadvertently bind your fellow songwriters into arrangements they never agreed to. Under principles of joint co-ownership, a single party’s grant—even one covering a non-exclusive license or covenant not to sue—can neutralize the enforcement rights of all contributors across an entire song. This means one publisher’s signature could effectively silence the ability of co-writers to pursue infringement claims against the AI platform.

Furthermore, beware of sweeping pre-signing releases. Some contracts attempt to bury a broad release of all claims arising before the agreement’s effective date. These clauses may seek to preemptively waive protections, such as those designed to protect individuals from releasing unknown claims—like California Civil Code Section 1542. If an AI company has been training on your catalog without permission for years, a backward-looking release could function as a contractual safe harbor, eliminating any leverage you have to seek compensation or an injunction.

The reality of the technology also poses unique risks. Terms surrounding training rights must be examined closely. If an agreement allows the AI company to embed your works into its model and make it nearly impossible to remove those works—even with “machine unlearning” techniques—you are effectively granting a right over your creative intelligence that you can never truly take back.

Finally, look beyond the royalty rate. Revenue pools are often riddled with layers of deductions—off-the-top fees, compute costs, and commissions—that carve down the total revenue significantly. Don’t be dazzled by a headline percentage; always model your actual payout by tracking every deduction. Don’t let an agreement define the end of the story; instead, read it like a litigator looking for remedies, not just a dealmaker looking for profit.