July 23, 2026. Warner took $500M from Suno. UMG signed with Udio. The major labels cashed out. The independent artists whose songs actually trained these models have been offered nothing. The asymmetry is the whole story.
In November 2025, Warner Music Group settled its lawsuit against Suno for $500 million and signed a licensing partnership. One month earlier, Universal Music Group had settled with Udio and announced a co-launched licensed AI music platform. Sony Music has settled with neither. Its fair-use cases against Suno and Udio are now pushed to 2027.
The math is stark. Two of the three major labels took the money and the licensing deals. One is still fighting. None of them represent independent artists.
The independent artists whose songs were scraped from YouTube and Spotify to train these models have received nothing. No settlement offers. No licensing deals. No seat at the table. Hagens Berman and Delgado Entertainment Law are running the class actions on their behalf, but those cases will take years.
In June 2026, Suno closed a $400 million Series D funding round. The company now has 2 million paid subscribers, generates 7 million songs per day, and reports annual recurring revenue of $300 million. Investors are betting that the Warner settlement provides sufficient cover. License the majors. Keep the training data secret. Outrun the class actions.
At the same time, Suno is fighting in court to conceal a single number: the total volume of audio files used to train its models. In a June 2026 filing, the company argued that revealing this figure would allow competitors to replicate and benchmark against Suno's models. Co-founder and CTO Georg Kucsko submitted a supporting declaration. The training data volume remains under seal.
The strategy is now visible. Raise money faster than the lawsuits can mature. Settle with the biggest plaintiffs. Fight discovery on everything else. By the time the class actions reach summary judgment, Suno will have either won, lost, or grown too large to care.
The Warner settlement gives Suno access to Warner's catalog. It gives Warner a royalty stream, board-level influence over Suno's AI likeness policies, and control over how Warner artists are represented in AI-generated music. UMG's deal with Udio follows a similar structure. These are not one-time payouts. They are ongoing licensing relationships.
For the majors, this is the rational play. Their catalogs are large enough to command nine-figure settlements. Their legal teams are deep enough to litigate for years if needed. And their business model, built on licensing, has a natural fit with AI training data. Pay us. Use our catalog. Everyone wins, except the people who were never at the table.
The independent artists got the Atlantic AI Watchdog.
In June 2026, Alex Reisner at The Atlantic published a searchable database of four music datasets used to train AI models. Two of them contain roughly 12 million and 9 million tracks, respectively. The other two contain more than 100,000 recordings each. Most were pulled from YouTube and Spotify using automated tools that violate the platforms' terms of service.
SZA checked the tool and found 238 of her tracks. She posted about it. DJ Sabrina the Teenage DJ found 22 of her songs. Independent artists across every genre checked and found their work. The Atlantic did not offer them a licensing deal. It offered them the truth.
If you want to know whether your music is in these datasets, you can search them yourself. We compiled the tools at beaumontandsheridan.com/resources/is-your-work-in-a-dataset. The Atlantic's AI Watchdog, Have I Been Trained for images, search tools for Books3 and Common Crawl. Find out what was taken. Document everything. And if you are eligible, join the class action.
Sony's fair-use cases are now the most important AI copyright litigation in the country. If a court rules that training on copyrighted recordings is not fair use, every AI music company must license training data or face liability. The Warner and UMG settlements become insurance policies, not windfalls. If the court accepts the fair use defense, the licensing leverage evaporates. The deals the majors signed will look like they got out early.
The Massachusetts court's June 30, 2026 scheduling order pushed dispositive motions to April 9, 2027. The Southern District of New York extended discovery to August 25, 2026. The summer 2026 fair-use ruling that the entire industry was waiting for is off the calendar. We will not have an answer until next year.
In the meantime, the asymmetry holds. The majors got paid. The indies got a search tool. And the models keep training.
We covered the training data story in depth on Episode 4 of The Echo. Listen here.
Beaumont & Sheridan is an information resource for individual creators. This post does not constitute legal advice.