In the mixtape era, DJs like DJ Clue and Funkmaster Flex worked out that the safest way to protect your IP was to put your name inside the record. Three seconds of your own voice over the intro, impossible to strip out without wrecking the track. No lawyers, no registration, no forms. A signature that travelled with the song wherever it went. If they don't sound familiar, I'm sure you've heard "DJ KHALED!" on the radio. But the rule was simple: bootleg or remix a song you didn't clear, you'd get sued. Diddy built "I'll Be Missing You" on Sting's "Every Breath You Take" without asking. The publishing went to Sting. All of it. Still does. Back then the price wasn't really the money. It was a foot in the door. The right bootleg could push you mainstream. Make you hot.
The AI labs run the same play in reverse. They sampled everything, every song, every book, every image, and cleared none of it. Pushing their models further and making them "hot". But when pressed for profit sharing, they called it their right: the information was readily available. Now they watermark their outputs so nobody can take from them, and their terms prohibit using those outputs to train a competing model. The tag and the watermark are the same instinct: protect the work at the level of the artefact. The difference is what was paid on the way in.
The DJs understood sampling came at a cost. The labs believe it's free for them to build their models and pass the cost to the users. Do we call it theft, or do we call it smart business?
🎵 Diddy, Faith Evans — I'll Be Missing You The most famous uncleared sample in history. From a producer who made a living off "borrowing" from others.
— Nicc