Two summers ago the record business was calling it theft. This summer it is calling it a strategic alliance.

In June 2024, the Recording Industry Association of America sued Suno and Udio on behalf of Sony, Universal, and Warner, alleging the AI music companies had trained on copyrighted recordings at industrial scale. Suno has since acknowledged its training set included essentially every music file of reasonable quality it could reach on the open internet. That is not a contested allegation anymore. It is a reported fact from the defendant.

Then the lawsuits started turning into paperwork of a different kind. Universal settled with Udio in October 2025. Warner settled with both companies in November 2025. On August 12, 2026, BMG announced a global alliance with Suno that both licenses its catalog going forward and resolves the prior unlicensed use. The next day, Spotify added the independent publisher Kobalt to the rightsholder pool behind its forthcoming tool for fan-made covers and remixes, following earlier deals with Universal and Merlin.

Six weeks of dealmaking. And in all of it, one question stayed unanswered in any document a working artist can read.

Global music publishing revenue reached $7.3 billion in 2025. That is the pot the AI deals attach to, alongside recorded music. When a company like Suno writes a check to settle past use and license future use, that check lands in a rightsholder's account. What happens next is governed by the same royalty math that has always governed it.

Here is the part that determines everything, and almost nobody in the coverage says it out loud: a settlement payment and a license fee are frequently classified as "other income" rather than as recording revenue. Under a standard record deal terms structure, other income is often shared on a different, less favorable basis than streams are, and in some agreements it is applied against an unrecouped balance before anything reaches an artist. An artist who is unrecouped, which describes most artists, may be technically included in a landmark AI settlement and receive zero dollars from it. Nothing improper has occurred. The contract simply did what it was written to do.

Tiffany Red, the Grammy-winning songwriter who founded the artist advocacy group the 100 Percenters, asked the question, speaking to NPR, in the most direct form available.

"Where's the money going, if they stole all of our music?"

It is not a rhetorical question. It has an answer, and the answer is a spreadsheet nobody has published.

What "opt-in" actually means

Every deal announced this month used the language of consent. Spotify’s global head of music, Charlie Hellman, described the Kobalt agreement as being “grounded in consent, credit, and compensation for the songwriters who take part.” BMG’s alliance with Suno is structured as opt-in, while Warner’s settlement gives artists the choice to approve or reject AI training involving their work.

Look more closely, however, and the decision-maker is often the rightsholder rather than the individual creator. Under an opt-in AI training model, Kobalt can approve a license where it administers a songwriter’s publishing. Likewise, a label can decide for a rapper whose master recordings it owns, even if the artist receives an opportunity to opt out later. Warner’s artist-level opt-out is therefore an exception, not the standard industry model.

Consent exists, but in many cases it sits one contractual level above the artist.

That makes master recording ownership a practical business variable, not merely a talking point. Artists who control their masters can negotiate an AI license, set a price, or refuse one. Those who do not may have limited influence beyond reviewing the contract and requesting an opt-out. The same month offered two contrasting examples. Megan Thee Stallion entered a distribution-only arrangement with Interscope through her own company while keeping her masters and publishing. Any AI deal involving that catalog therefore runs through her. T-Pain instead sold his catalog outright, on terms he explained publicly, giving the buyer control over future licensing decisions. Neither strategy is inherently right or wrong. Artists assessing similar deals should identify who owns the masters, who administers the publishing, and who holds AI approval rights before signing.

The supply problem underneath the demand problem

As legal battles over AI-generated music move into the marketplace, the volume of releases is reshaping the economics of streaming. Suno surpassed two million paid subscribers in February 2026. On Deezer, AI-generated tracks now represent more than half of daily uploads. The figure refers to uploads, not plays, but that difference matters to musicians whose income depends on being discovered.

This is why the issue is also a streaming royalties story. Most streaming services distribute revenue from a shared pool based on each track’s proportion of total plays. Every new release competes for attention and claims a potential share of that pool. Entertainment lawyer Krystle Delgado, a plaintiff in a class action against AI companies, summarized the concern by noting that human and AI-generated music enter the same catalog. Without effective screening, platforms have limited ways to distinguish between them.

Some services are introducing stricter controls. On August 13, Beatport banned tracks that are entirely or mostly AI-generated. The platform uses detection technology from fraud analytics firm Beatdapp to flag submissions during intake. AI-assisted tracks made with meaningful human involvement remain allowed, but they are tagged for curator review.

Beatport’s user survey also points to a financial concern. Sixty percent of surveyed DJs said they would not play an AI-generated track, while only 8 percent supported the idea. Another 13 percent would consider playing one if artists received fair compensation. The findings suggest that opposition is driven less by sound quality alone than by questions about attribution, competition, and how streaming revenue is shared.

What remains unsettled in AI music licensing

Sony's litigation against AI companies remains active, while a separate dispute is testing how existing settlements affect musicians. In June 2026, the American Federation of Musicians sued Universal Music Group and Warner Music Group, alleging that the labels allowed member recordings to be used without proper licenses in AI products. Both companies moved to dismiss the case.

The dispute highlights a gap in the music industry's AI agreements. Major labels may have settled or negotiated terms with AI companies, but those deals do not automatically resolve the rights of session players, performers, or other contributors whose recordings are included in the affected catalogs. The union's lawsuit involves parties that are broadly aligned on the need to protect music rights, yet they still disagree over how artists should be compensated.

In practical terms, the settlements addressed the relationship between AI companies and music rightsholders. They did not settle the relationship between those rightsholders and the people who performed on the recordings. That second negotiation is only beginning and will likely be decided contract by contract.

What working artists should do this month

Review the two clauses that may control AI payments

Locate your recording, publishing, administration, and licensing agreements. Start with the clause covering new, future, or unknown technologies, then review the language defining other income, net receipts, or ancillary revenue. These provisions may determine whether an AI licensing payment reaches you and how much the label, publisher, or administrator can retain.

If your contract predates generative AI and does not mention machine learning, do not assume that silence settles the issue. Ambiguous language can create negotiating leverage, but only if you identify it and raise the question before signing an amendment, renewal, or new license. Consider having an entertainment lawyer review the relevant provisions.

Address machine-training rights before signing

Be specific about music catalog rights in every new agreement. The central issue in 2026 is not only who owns the master recording. It is also who can license that recording for machine training, whether that permission is included in existing rights, and whether it can be excluded or separately priced. Artists with legal representation are already seeking those carve-outs.

Review your publishing splits as well. Spotify remix arrangements and similar AI-related deals may involve publishing rights rather than recordings. A writer with an administration agreement could be included without a separate negotiation, while a self-administered writer may need to decide whether to participate independently.

The broader lesson is that artist leverage increasingly depends on paperwork, not popularity alone. Your catalog may provide your strongest negotiating position, but only if you understand what you have licensed. Suno's public position is that people will continue making the best music. The unresolved question is who gets paid when AI systems learn from that music. The answer is being written into contracts now, often in language artists may never see unless they ask for it.