In August 2026, millions of Twitch streamers discovered that their broadcasts, video-on-demand recordings, clips, chat logs, and channel images had, by default, been made available to train Amazon's generative artificial intelligence models. A new toggle in the platform's Security and Privacy settings can stop that — but only if users find it and switch it off themselves. The opt-out is narrow in scope but significant in principle. According to a Twitch support page, deactivating the "Training for Generative AI" setting prevents a streamer's content from being used in the future training of any Amazon model designed to generate or synthesise text, audio, images, or video.

What the opt-out does not do is remove a creator's data from every Amazon process: the company is explicit that content may still be used for other purposes described in its Privacy Notice, including recommendation algorithms, real-time sponsorship tools, and its AutoMod safety system. There is a further complication for anyone who participates in another streamer's chat. Even if a viewer has opted out on their own channel, Twitch has confirmed that the channel owner's preference governs whether that chat data can be used for training — meaning a single person's opt-out provides no protection the moment they type a message somewhere else. This asymmetry matters because Twitch's global audience, concentrated heavily in North America, Latin America, and Europe, routinely interacts across thousands of channels daily.