A Twitch streamer has filed a proposed class action against Twitch and its parent company, Amazon, claiming that creators’ broadcasts were used to train generative AI without consent or payment, Eurogamer reported, citing Courthouse News.
Plaintiff Warren Pandiscia, a creator with about 1,000 followers, filed a 37-page complaint in the U.S. District Court for the Northern District of California. It alleges that Twitch and Amazon copied large amounts of streamer footage without proper licenses, breached creators’ implied agreements with the platform and violated state unfair competition law. Pandiscia says he opted out as soon as the setting existed and argues the companies had been using streamer content for AI training as far back as 2024, before any disclosure or opt-out. Neither company had responded to Courthouse News’ request for comment.
An Opt-Out That Arrived Late
Twitch said on Aug. 12 that it would let users opt out of having their “streams, VODs, clips, stream chats, and pictures and text” used to train Amazon’s generative AI content models, Ars Technica reported. Users are opted in by default, and the opt-out is not retroactive. It is also tied to individual channels, so a streamer who opts out can still appear in training data through another channel that has not.
Amazon’s use of Twitch content was not entirely secret. Asked at a 2024 event hosted by The Information whether Amazon trains AI on Twitch, then-chief monetization officer Mike Minton said, “Yeah, for sure.” Ars noted that online discussion before the announcement still showed widespread uncertainty about the practice.
Minton, now Twitch’s chief product officer, was blunt about the default setting. “If this was opt-in, nobody would opt in. That’s honestly the answer,” he said, according to TechCrunch. Eurogamer quoted him adding, “I know this is not a fan favourite, I know it’s very upsetting to the community, but this is where we are.”
What the Case Tests
The suit adds to a growing list of disputes over whether platforms can use user-generated content to train AI under existing terms of service. Twitch’s archive of live video, audio and chat is valuable training material, and the case will test whether a default-on opt-out added years after training began is enough. For platforms that host creator content, the clearer path is explicit disclosure before training starts and a consent setting that covers past uploads, not only future ones.
WebProNews is an iEntry Publication