01What happened
The story, straight
SZA discovered that 238 of her songs, including what she suspects are unreleased tracks, were included in data sets used to train AI music generators. The discovery came via a new detection tool launched last week by The Atlantic, built by researcher Alex Reisner, which lets artists search whether their music appears in four major AI training corpora encompassing over 21 million songs. Producer Kenneth Blume also publicly condemned the practice after checking the tool. SZA posted on Instagram: 'Jus checked and music AI has trained off 238 of my songs. I'm certain some unreleased. If your a musician and you support this degenerate shit? Your disgusting and there's NOTHING YOU COULD EVER SAY TO ME TO MAKE THIS OKAY.' The tool draws on data sets that include catalogs from Bad Bunny, Taylor Swift, Beyoncé, and independent artists alike.
sza ran her name through the atlantic's new AI music training detection tool and found 238 of her songs already scraped — some unreleased, she thinks. she went off on instagram about it. producer kenneth blume also checked and publicly condemned the whole thing. the tool was built by researcher alex reisner and scans four data sets totaling 21M+ songs that AI developers are using. bad bunny, taylor swift, beyoncé, tons of indie artists — all in there.
02Spread timeline
Where it actually started
03Source receipts
Every claim, linked
04What's solid, what isn't
What's solid and what isn't
- The Atlantic launched an AI training data detection tool built by researcher Alex Reisner
- The tool covers four data sets encompassing over 21 million songs
- SZA found 238 of her songs in the data sets, including suspected unreleased material
- Kenneth Blume also publicly condemned the use of his songs in AI training
- Whether unreleased SZA tracks are confirmed in the data sets or if this is her own assessment
- The exact scope of Kenneth Blume's discovery — how many songs were found
- Additional artists expected to check the tool and react publicly in coming days
- Potential legal action from artists whose work was included without consent
05Why it matters
The editorial take
The Atlantic's new detection tool has made the scale of AI music training tangible for artists for the first time — and the reaction is immediate, furious, and from the highest-profile names in music. This follows a broader pattern of artists discovering their work was used without consent to train generative AI, and it lands as multiple lawsuits over AI training data continue working through courts.
this is the first time artists can actually *see* the scale of the scraping in a searchable way, and the reaction is exactly what you'd expect. 238 songs from one artist alone. we're past 'is this happening' — the conversation is now 'what are we doing about it.' expect more names, more anger, more lawsuits.
