Twitch opts streamers into Amazon AI training by default – you have to opt out

What this is about

Twitch, the Amazon-owned livestreaming platform, introduced a setting called "Training for Generative AI" in mid-August 2026. It is turned on by default and allows content from channels to be used to train generative AI models at Amazon. Generative AI models create new content on demand, such as text, images or speech. If you don't want that, you have to actively opt out. According to TechCrunch, the community reacted quickly and with fierce criticism.

What is affected and how to opt out

According to TechTimes, the permission covers livestreams, archived videos, clips, chat logs and channel images. Twitch's FAQ gives one example: audio could help improve speech recognition models (speech-to-text), which would make captions better on Twitch and across Amazon.

Here's how to turn it off: click your profile picture, then Settings (not the Creator Dashboard), open the "Security and Privacy" tab, scroll down and switch off "Training for Generative AI." According to Twitch's FAQ, AI-powered features such as captions and safety tools are not affected. According to a post by reporter Zach Bussey embedded by Shacknews, opting out also applies to existing content, so it won't be used for future models.

The head of product's admission

In a livestream with nearly 3,000 viewers, head of product Mike Minton and community chief Mary Kish faced the criticism. Asked why there is no opt-in, Minton answered, according to TechCrunch: "If it were opt-in, nobody would agree. That's honestly the answer." Kish defended the move, saying the very existence of an opt-out option was a response to what the community wanted. She also pointed out that Meta likewise uses public content from its platforms for AI training.

The communication is under fire too: Twitch described the change on X as a new setting that lets you object to the use. According to TechCrunch and TechTimes, this confused many streamers because it sounded as if a protection had been added, when the switch was in fact already active. TechTimes quotes PC Gamer, saying a Twitch reply to Bussey even claimed that Twitch does not train GenAI models on streamer content, which contradicts the FAQ.

An unclear past

Asked whether earlier videos had already been used, Minton said he didn't know, because he didn't know what Amazon had used in model training. TechTimes also points to a 2024 interview in which Minton told The Information that Twitch content was already being used "in prototyping, not at production scale" for AI development. According to TechTimes, these are signs that the use began earlier than the notice to creators.

According to the reports, the value of the data lies in its nature: streams deliver video, voice and chat at once, unscripted and over many hours. TechTimes (figures as given by the outlet) cites roughly 7.3 million monthly active streamers and 240 million users. According to Engadget, Minton confirmed that Twitch already transcribes nearly all audio, for example for summaries and moderation.

Reactions, legal issues and the class action

TechTimes reports a UserVoice thread with about 14,930 votes for making AI features opt-in. Games consultant Mike Futter raised the question of what happens if a creator streams a copyrighted game and has agreed to AI use. In his view, studios would probably consult their lawyers. TechTimes also classes the approach as a "dark pattern" and points to FTC guidance and California law. That is the outlet's assessment, not a court finding.

According to Shacknews, a class action has since been filed in San Francisco (plaintiff: Warren Pandiscia; reported by Courthouse News Service). It accuses Twitch and Amazon of using streams without a license or consent as a mass dataset for AI products, and argues the systems could never obtain the consent of everyone involved. The plaintiff says he learned of the option from a journalist, not from Twitch. The allegations are so far only claims by the plaintiff's side.

Our take

If you stream, the practical advice is simple: check the switch and turn it off if you want to. The case shows how platforms secure user data for AI through default settings, much like Meta. The manager's candid statement is unusual. What remains open is what Amazon did with the data before August and whether opting out affects models that have already been trained. Twitch hasn't answered that. How the lawsuit ends is also unclear. And the question doesn't only concern streamers, but also game studios whose content appears in streams.