Lawsuit alleges Apple scraped YouTube
A class-action complaint claims Apple scraped YouTube videos to train AI models without creators’ permission, bringing creator copyright and AI-training rights into sharp focus. The case is early and contested, but it already changes negotiation leverage: creators and brands should now be explicit about whether licensed assets may be used for AI training or synthetic model development. (voi.id)
Three YouTube channels sued Apple in federal court in California after alleging Apple used their videos to train artificial intelligence systems without permission. The plaintiffs are h3h3Productions, H3 Podcast, MrShortGame Golf, and Golfholics, and the complaint was reported on April 6 and April 7, 2026. (macrumors.com) The lawsuit says Apple did not just watch public videos like a person with a browser. It says Apple bypassed YouTube’s anti-scraping barriers and copied video files at industrial scale, which the creators frame as a violation of the Digital Millennium Copyright Act, the United States law that bans dodging access controls on copyrighted works. (pcmag.com) The number at the center of the case is 173,536 videos from roughly 48,000 channels. Some reports on the complaint also say the broader dataset linked to the dispute contains about 70 million video-caption pairs built from 3.8 million longer source videos. (msn.com) (snap-research.github.io) That dataset is called Panda-70M, and it works like a giant chopped-up library of internet video. Researchers took millions of long videos, split them into shorter clips, and attached text captions so a model could learn what moving images look like and how to describe them. (openaccess.thecvf.com) Panda-70M was introduced in a 2024 computer vision paper by researchers affiliated with Snap and academic partners, not by Apple. But the lawsuit points to Apple’s later research paper on video generation, called STIV, which lists Panda-70M among its training datasets. (openaccess.thecvf.com) (machinelearning.apple.com) That link is why Apple is in court now. The creators argue Apple’s own paper is the breadcrumb trail showing that clips derived from YouTube ended up inside Apple’s model-training pipeline. (appleinsider.com) Apple’s STIV paper is about making short videos from text prompts and still images. In plain English, it is the kind of system you build by feeding a model huge amounts of labeled video so it can learn patterns like “a golfer swings” or “a person talks into a microphone.” (machinelearning.apple.com) The legal fight is not over whether YouTube exists in public. It is over whether a company can turn creators’ uploads into training fuel, especially if getting that fuel required technical workarounds that creators and the platform never approved. (engadget.com) The case is still at the complaint stage, which means the court has not decided whether Apple actually broke the law. But even before any ruling, the filing gives creators, studios, and brands a new checklist item: if you license video, spell out whether that license does or does not include artificial intelligence training and synthetic media development. (9to5mac.com)
