Streamer sues Twitch and Amazon over AI training on a decade of livestreams

A Twitch streamer has sued the platform and its parent company, Amazon, alleging that years of his livestreams, video-on-demand archives and chat logs were copied without permission to train Amazon's generative artificial intelligence systems. The suit, filed in federal court in San Francisco and first reported by Courthouse News Service, is among the first class actions to target Twitch directly over the practice, and it lands ten days after the company quietly gave creators a way to opt out of it.

Warren Pandiscia, a streamer from Fairfield County, Connecticut, who has broadcast Lego builds and video games on Twitch for nearly a decade, filed the 37-page complaint on Aug. 20 in the U.S. District Court for the Northern District of California, seeking to represent every Twitch creator whose content was used the same way. The case, Pandiscia v. Twitch Interactive Inc. et al., No. 3:26-cv-08721, has been assigned to Magistrate Judge Lisa J. Cisneros, with an initial case-management conference set for Nov. 19.

What the complaint alleges

The complaint accuses Twitch and Amazon of accessing streams, clips, VODs and chat transcripts dating back to 2024 to build training data for Amazon's commercial AI products, without seeking a license or creators' consent. It frames the conduct as a breach of both the express terms of Twitch's creator agreements and an implied contract arising from a decade of platform practice, along with unjust enrichment and violations of California's Unfair Competition Law. "Because Amazon AI products are commercialized, Amazon had an overwhelming incentive to acquire training data on an unprecedented scale," the filing states, adding that rather than negotiate licenses, "defendants accessed the Twitch streams and videos to utilize them as a massive dataset necessary to fuel Amazon's AI products."

Pandiscia is asking the court for damages, restitution and disgorgement of any profits tied to the training data, as well as an injunction that would require Twitch and Amazon to strip creators' material out of existing AI datasets. Twitch has roughly 7 million channels that go live in a typical month, according to industry tracking figures, though only a fraction earn any revenue from the platform — meaning the potential class, if certified, could be large even by the standards of recent AI-training litigation. Pandiscia's own channel, the complaint states, has drawn more than 900 followers and over 1,000 hours of broadcast footage across nearly ten years on the platform, none of which he was asked to license for AI development.

A default nobody asked for

The lawsuit follows directly from a change Twitch made with little fanfare on Aug. 12, when it added a "Training for Generative AI" toggle to the security and privacy section of account settings. The setting, described on Twitch's own help pages, covers a channel's streams, VODs, clips, chat logs and the images and text on a channel page. Twitch's support account confirmed the change on social media the same day, after a streaming reporter spotted it first. There was no advance notice to creators — no email, no in-app pop-up — and the toggle arrived switched on for every existing account, meaning years of past content had already been eligible for training before anyone could decline.

Twitch's chief product officer, Mike Minton, addressed the design choice directly during the company's "Patch Notes" livestream on Aug. 13, as backlash built in the chat. Asked why the feature defaulted to enrollment rather than requiring creators to opt in, he was unusually blunt.

"If it was opt in, nobody would opt in. That's honestly the answer. So it's going to be on by default."

Minton added that "almost every content service in the world" makes similar data uses opt-out rather than opt-in, a framing that did little to quiet criticism from creators who said they learned only after the fact that a decade of footage had already been available to Amazon's AI teams. Coverage of the remarks by outlets including PC Gamer and the Video Games Chronicle spread quickly among creators already unsettled by the rollout. Even for those who now switch the setting off, the help documentation notes an exception: chat messages posted in someone else's stream are still eligible for training unless that other streamer has separately opted out.

Part of a wider pattern

Pandiscia's suit is the latest in a run of creator lawsuits over unlicensed AI training that has moved through federal courts this year. In April, a group of YouTube channels led by the podcast Ted Entertainment sued Apple, alleging the company scraped more than 170,000 videos to help build machine-learning models; Apple has argued in filings that the videos were public and that its access was lawful. A separate group of YouTubers sued Snap in January over similar claims. None of those cases has yet reached a jury, and legal specialists say the theories at issue — breach of implied contract and state unfair-competition law, rather than copyright infringement outright — remain largely untested against large technology platforms.

For working streamers, the stakes are practical rather than abstract. Most Twitch creators rely on subscriptions, donations and ad revenue built on an audience relationship, and many say they never anticipated their broadcasts becoming raw material for AI systems that could, in theory, generate synthetic video, voice or text in a similar style. Pandiscia's complaint argues that this unanticipated use is precisely what makes it a breach of the understanding creators had with the platform when they agreed to stream there. As Engadget noted in its coverage of the filing, the case adds Twitch to a growing list of platforms — alongside YouTube, Snap and others — now facing creator-driven claims that AI training amounts to uncompensated use of their work.

What happens next

Twitch and Amazon have not filed a response to the complaint, and neither company has commented publicly on the litigation specifically. Under the scheduling order issued by the court, the two companies have until early autumn to respond, and a case-management statement is due Nov. 12 ahead of the first conference a week later. Any motion to dismiss, and any fight over class certification, would likely stretch well into 2027. In the meantime, the underlying setting remains live on Twitch: creators who want their future content excluded from Amazon's model training still have to find the toggle and switch it off themselves.