Twitch users now have a twitch AI training opt out setting that blocks their channel material from future training of Amazon generative AI content models. The control is on by default, according to contemporaneous reporting, so creators who do not want their work used need to change it themselves.
The setting is described as applying to future training. It does not say that any past training use is reversed or removed. That distinction is doing a lot of work here, because the public record still does not establish when Amazon began using Twitch material, which models received it, or the full scope of any historical use.
Twitch’s support documentation says the covered material includes streams, video-on-demand archives, clips, stream chats, and pictures and text posted on a user’s channel. The models covered are those intended to generate or synthesize text, audio, images, or video, according to reports by Ars Technica and The Verge.
How do you opt out of Twitch AI training?
Open Twitch account settings, rather than the Creator Dashboard, select Security and Privacy, find Training for Generative AI, and switch the control off. TechCrunch and other outlets reported that path on August 12.
The setting is narrower than a general Amazon or Twitch data-use ban. Twitch says disabling it does not turn off other AI-supported functions covered by its privacy notice. Those can include captions, recommendations, sponsorship assistance and community-safety tools such as AutoMod, The Verge reported. In other words, the toggle addresses training of generative content models, not every automated system operating around the service.
What does the Twitch AI training opt out not cover?
Chat creates an awkward exception. If a user who has opted out writes in another streamer’s chat, the other channel owner’s setting determines whether that message may be used for training, Twitch says. A personal preference therefore does not follow a user into every channel they visit.
Twitch’s own example of permitted training explains the incentive: audio from a channel could help improve speech-to-text systems, including captioning on Twitch and elsewhere across Amazon. That is a company description of a potential use, not a complete account of every model or dataset involved.
Amazon bought Twitch in 2014. In 2024, then-Twitch chief monetization officer Mike Minton answered “Yeah, for sure” when asked whether Twitch had a role in training Amazon AI, Ars Technica reported. He added qualifications about user trust and privacy rules. More recently, Minton said on an official Twitch stream that he did not know what Amazon had used in model training when asked whether videos had already been used, according to TechCrunch.
That leaves the history unresolved. The new control gives channel owners a choice about future generative-AI training, but it is not a receipt for what Amazon has already done with the platform’s archive.
This story draws on original reporting from Ars Technica.
