Class Action Accuses Amazon and Twitch of Training AI on Streamers’ Content

Amazon and Twitch have been hit with a class action in the US District Court for the Northern District of California. The case was filed last Thursday by US streamer Warren Pandiscia, who says the companies used millions of Twitch videos to train generative AI without licences or payment.

The complaint, which runs 37 pages, says the platforms pulled in streams and videos as a huge training set instead of asking permission. Pandiscia’s argument goes further than the usual creator gripe. He says Amazon’s setup makes real consent impossible, because a stream can include guests, chat viewers and other people who never owned the channel in the first place.

The dispute sharpened after a Twitch change in mid-August. From then on, Twitch content can be used to train Amazon’s generative models, including live streams, VODs, clips, highlights, stream chats, channel images and text. Twitch says the point is things like better speech-to-text and captions, with those improvements potentially flowing into other Amazon services. Amazon, meanwhile, is building generative models for text, audio, images and video.

The system is opt-out, not opt-in. Channels are included unless the owner turns off the “Training for generative AI” setting under security and privacy. That only covers that channel, though. If someone appears in a stream whose owner leaves training enabled, their own objection does not stop anything. Twitch chief product officer Mike Minton also added fuel when he said on an official patch notes livestream that, if it were opt-in, nobody would opt in.

Pandiscia also says Twitch analysed video material going back to 2024, before the opt-out option existed at all. That leaves the company facing a messy question: can a later setting clean up earlier use? The suit brings claims including breach of express contract, breach of implied contract, unjust enrichment and unfair business practices, and it seeks damages, restitution, disgorgement of profits and an order to remove the material from training datasets. Amazon and Twitch have not publicly responded.

The case lands in the middle of a larger run of AI-training lawsuits. A YouTuber has already sued Nvidia over alleged use of his content, Anthropic paid 1.5 billion dollars to settle a books case, and Meta won one over books used for training. But this one is nastier for platforms, because it says the platform itself is feeding user content to its parent company under a default setting and calling that consent. That is not a great look, even by Silicon Valley standards.