20 Million Songs Just Got Exposed as AI Fuel
Last week, something happened that every musician needs to see.
The Atlantic’s AI Watchdog project published four searchable databases containing the songs used to train AI music models. Not a vague allegation. Not an industry rumor. A searchable, public tool where you can type in your artist name and find out if your music was fed into the machine.
The numbers are staggering. The largest dataset contains 12 million tracks. A second holds 9 million. Two smaller sets each exceed 100,000 songs. That’s over 20 million tracks in total — and those are just the four datasets one journalist could find.
Within 48 hours, musicians started posting their results. The reactions ranged from stunned to furious.
The Week Musicians Learned the Truth
The investigation, led by Atlantic staff writer Alex Reisner, didn’t just drop a number. It dropped a searchable interface. And that changed everything.
Breakcore producer Sophiaaaahjkl;8901 posted about the searchable database on social media, prompting numerous other musicians to share screenshots of their own search results. Her discovery was personal: “Suno and Udio [have] used 138 of my songs across two of their datasets,” the artist wrote. “This is almost my entire catalogue of music. It’s just about everything I’ve released from 2017 to 2024.”
She has about 10,000 followers. Not a major label artist. Not a chart-topper. Just an independent musician whose entire body of work was apparently absorbed into training data without her knowledge.
Producer Vince Valholla, head of Valholla Records, posted a video on X: “Late last night I found out over 100+ songs from our catalogue were used to train AI models. To be honest, until the major labels go through their lawsuits, there’s no way for artists or labels to fight back. They literally scraped the best songs from our catalogue.”
Backxwash, Titus Andronicus, Tre Mission, Lunice, and DJ Sabrina the Teenage DJ are among the musicians who have expressed disdain for finding their music within the databases.
What The Atlantic Actually Found
Let’s break down the four datasets, because the details matter.
The tool draws upon four datasets. One dataset contains 12 million tracks, while another contains 9 million tracks. The two other datasets both contain over 100,000 tracks.
The presence of hits from commercially prominent acts won’t come as a surprise — roughly 300 Beatles tracks are in each of the two biggest datasets, as are hundreds of songs apiece from Taylor Swift, ABBA, Snoop Dogg, and Michael Jackson.
But here’s the part that’s genuinely chilling for indie artists: the datasets don’t discriminate. Numerous artists are now discovering that their work appears in datasets assembled for model training or research. Reisner names major catalogs and less visible musicians in the same data field.
In three of the datasets, songs were distributed via links to YouTube and Spotify, which are often accessed using “tools that automate the job, some of which allow developers to bypass logins, advertisements and mechanisms that might earn money or subscribers for creators.” In other words: stream-ripping at industrial scale.
Companies like Google and Stability have openly utilized these datasets, but due to the AI industry’s opaque mechanics, it’s currently unclear who else has used them, and how many others like them may exist.
What the Databases Don’t Prove
This is important, and responsible coverage demands saying it clearly.
As of June 2026, the investigation does not name a single AI company as having trained on a specific dataset. Music Ally specifically noted that the article “does not specify which particular AI companies have utilized them.” The datasets are identifiable; their downstream use is not pinned to a named product in the reporting.
A track appearing in the tool doesn’t definitively mean it was used to train AI, but due to the existence of other datasets, a track’s absence is also not evidence it hasn’t been used.
So the databases are evidence of exposure, not a smoking gun. But paired with court filings? They’re devastating.
The Legal Firestorm This Feeds
The Atlantic’s databases land right as the AI music copyright war reaches its climax.
Suno is fighting all claims on fair use grounds in the District of Massachusetts, with a key summary judgment hearing scheduled for July 2026. That hearing could define whether training AI models on copyrighted music is legal in the United States. Full stop.
Here’s the current scorecard:
UMG settled with Udio in October 2025, creating a per-generation licensing template.
Warner Music Group reached its own settlement and licensing deal with Udio in November 2025, and days later became the first major to settle with Suno. The Warner–Suno agreement also saw Suno acquire the concert-discovery platform Songkick from WMG.
Sony Music Entertainment is the last major record label still in court.
A ruling against Suno on fair use would force every AI music company to license training data or shut down. A ruling for Suno would gut the labels’ negotiating leverage and reset the licensing market overnight.
And now? For artists, searchable proof of which songs trained a model is the hard evidence these cases have lacked. It strengthens the labels’ cases against Suno and Udio, and it raises the pressure on platforms that still will not disclose what their own tools were trained on.
Meanwhile, Suno has raised a $400 million Series D round at a $5.4 billion valuation — more than double its previous round — even as it fights copyright lawsuits from the world’s biggest record labels. Investors are betting billions that fair use wins. The music industry is betting its future that it doesn’t.
The Indie Artist Problem
The major labels have legal teams, settlement deals, and licensing revenue flowing. But independent artists? They’re in a very different position.
The Nguyen class-action complaint alleges that Suno’s training data included over 40 million tracks, of which at least 60% came from independent artists. If accurate, the majority of Suno’s training data was created by artists who have no licensing agreement and no pathway to compensation.
A study commissioned by CISAC estimated that generative AI could take 24% of music creators’ revenues by 2028 — a cumulative loss of €10 billion ($10.5 billion) for creators between 2023 and 2028.
That’s not abstract. At the level of an individual act, the instrumental duo The American Dollar alleged in a May 2026 lawsuit that Suno had cut its licensing revenue by nearly 80%.
What This Means for Your Visual Strategy
Here’s where the conversation pivots from outrage to action. Because while the legal battles over AI music generation are raging, AI music videos occupy a fundamentally different space.
When you create an AI music video, you’re using AI as a visual tool for your original music. The music is yours. The creative direction is yours. The AI generates visuals to serve your song — it doesn’t replace your artistic output. It amplifies it.
This is the crucial distinction that matters in 2026: there’s a world of difference between an AI that generates music from scraped training data and an AI that generates visuals to accompany your original music. The Atlantic’s investigation is about the former. Tools like OneMoreShot.ai are the latter.
If anything, the training data revelations make a strong case for why independent musicians should double down on visual content. Your music is being scraped, cloned, and fed into models that generate competing output. Your visual identity — your aesthetic, your brand, the way your music looks and feels on screen — is one of the few things AI music generators can’t replicate.
Whether you’re making hip-hop visuals, EDM content, or indie music videos, the visual layer is where you differentiate yourself from the flood.
The Bigger Picture: Proof Is the New Currency
AI music creators are entering the proof era. Not just proof that a song sounds good. Proof of process. Proof of rights awareness. Proof of human direction. Proof of metadata. Proof that the creator is not just flooding platforms with disposable AI output.
The Atlantic’s databases accelerate this shift dramatically. We’re moving into a world where:
-
Platforms demand transparency. Apple has moved toward AI transparency tags. Deezer has reported massive AI-generated upload volume and uses detection/tagging to identify AI-generated tracks. Spotify has been working on AI disclosure standards through broader industry metadata efforts.
-
AI-generated music gets quarantined. Deezer keeps fully AI-generated songs out of algorithmic recommendations and editorial playlists to prevent them from gaining more visibility than music made by real people.
-
The flood keeps growing. Deezer reported that by November 2025 it was receiving over 50,000 fully AI-generated tracks each day, making up around one-third of all new deliveries to the platform. Spotify said it removed more than 75 million spam or low-quality tracks over a 12-month period.
In this environment, being a real musician with real visual content attached to your real songs isn’t just an artistic choice — it’s a competitive strategy. A complete visual package tells algorithms, playlist curators, and fans that you’re a legitimate artist, not a content farm.
Check out our complete guide to AI music videos for a deep dive on building that visual identity.
How to Check If Your Music Was Scraped
You should absolutely look yourself up. Here’s what to know:
- Visit The Atlantic’s AI Watchdog tool. Search by artist name or song title.
- Check all four databases. A song can appear in one, multiple, or none.
- Screenshot everything. If you find your music, document it. These records could become evidence.
- Understand the limits. Presence in a dataset doesn’t prove a specific company trained on your track. Absence doesn’t prove they didn’t.
- Consider your options. Artists may use the database to inform legal claims, licensing negotiations, or public campaigns. Researchers may use it to map connections between datasets and published AI work.
Australia’s official music copyright team, APRA AMCOS, has also announced it will be launching an investigation into The Atlantic’s findings. Expect more collecting societies to follow.
The Road Ahead
The next month is going to be wild. The Suno fair-use hearing in July 2026 is arguably the most important moment in music copyright since Napster. A federal court is weeks away from a hearing that may determine whether training an AI on copyrighted recordings is legal under U.S. copyright law. The answer will matter not just to Sony Music and the two AI startups it is still fighting, but to every company that has ever built a generative model on data it did not license.
Mat Dryhurst responded to the Atlantic piece with a useful correction from inside the artist and AI debate. His point is not that the investigation is wrong. It is that musicians often bundle several different problems under the idea of training data: market competition, payment for data, impersonation, and a deeper wish that AI culture would disappear. Those concerns touch each other, but they do not lead to the same remedy.
That’s a genuinely important observation. The solutions for “they used my song to train a model” and “AI-generated music is flooding Spotify” are different problems requiring different fixes.
But for working musicians right now, the practical move is clear: own your visual identity before the machines eat that too.
AI video tools are creating new possibilities for every genre — from lo-fi aesthetics to Latin visual storytelling. The key is using AI as your creative tool, not having AI use you as its training data.
Make Your Music Visible
The training data wars are just getting started. But while the lawyers fight over who owns the data that trained the models, you can be making music videos that put your artistry front and center.
OneMoreShot.ai lets you create stunning, beat-synced music videos from your original tracks in minutes. No crew. No five-figure budget. No scraped training data controversy. Just your music, your vision, and visuals that make people hit play.
Because in a world where 20 million songs just got exposed as training fuel, the best thing you can do is make your music impossible to ignore.