Twitch adds opt-out for Amazon AI training, enabled by default
Twitch has updated its account settings to let streamers opt out of having their posts, streams and videos used to train the AI models of its parent company, Amazon. The toggle sits under Settings, then Security and Privacy, then Generative AI Training, reachable from the account avatar on the website or the mobile app. It is enabled by default, meaning a streamer's content is used for AI training unless the streamer turns it off. The text next to the toggle also warns that disabling it does not stop Twitch and Amazon from using a channel's content for other purposes described in Twitch's Privacy Notice, including AI-powered features meant to help streamers grow and monetize, such as real-time sponsorship assistance, viewer discovery through recommendations, and community-safety tools like AutoMod.
The update has reassured some creators but has also exposed how little Twitch had disclosed before now. It remains unclear exactly when the platform began feeding user posts, streams and videos into Amazon's AI training pipeline. In a forum dedicated to the topic, more than 16,000 creators expressed opposition to their content being used by default for this purpose, a practice that only came to light once the new setting appeared.
The backlash followed a livestream in which Mary Kish, Twitch's head of community, explained the changes and acknowledged that they would provoke a negative reaction. Mike Minton, Twitch's head of product, gave what he called "a candid response," saying that keeping the AI-training option enabled by default was necessary because otherwise "no one would participate." Minton also said the practice is not unique to Twitch and that he considers it reasonable to assume other AI developers are extracting content from Twitch and similar services for training, whether or not they have permission: "I don't know for sure, but I think it's quite reasonable to assume that almost any publicly available content is used to train models in one way or another, with or without permission. So I think we also need to acknowledge that there's a lot here that's beyond even our direct control."
Twitch's Terms of Service, in effect since March 2024, already granted Twitch and its sublicensees the right to use, reproduce, modify, adapt, distribute and create derivative works from user content, but until now had not explicitly stated that this content could be used to train generative AI models.
The episode is framed as one symptom of a broader shortage of high-quality training data. OpenAI has struck agreements with publishers, including WIRED's corporate parent Condé Nast, to license some of their content, but usable data sources are shrinking as demand grows. Meta has been using posts and images that people share on Facebook and Instagram to train its models, and it recently came to light that the company was also using its own employees' activity and screenshots captured by users of its smart glasses for the same purpose. Similar practices have reportedly involved Google and YouTube. High-quality training data, the piece argues, has become as strategic a resource as chips, memory and electricity, much of it drawn from content users generate in exchange for free access to everyday digital services.
Key facts
- Twitch added an opt-out for AI training under Settings, Security and Privacy, Generative AI Training, but the toggle is enabled by default, so streamers must act to stop their content from being used.
- More than 16,000 creators expressed opposition in a dedicated forum after learning the default-on training practice, which only surfaced once the new setting appeared.
- Twitch's head of product Mike Minton said the default-on setting was necessary because otherwise 'no one would participate,' and called broad scraping of public content for AI training likely across the industry.
- Twitch's Terms of Service have granted Twitch and its sublicensees broad content-use rights since March 2024, but did not explicitly mention generative AI training until the new setting was introduced.
- The article ties the episode to reported AI-training uses of user content at Meta (Facebook, Instagram, employee activity, smart-glasses screenshots) and at Google and YouTube.
Why it matters
The setting is the first time Twitch has explicitly acknowledged using streamer content to train Amazon's AI models, and it surfaces a question the company had left unanswered: how long this had already been happening before anyone could opt out. It lands amid a broader industry shortage of high-quality training data, which the piece treats as a resource on par with chips, memory and electricity, one that companies increasingly meet by drawing on content their own users generate.
Who it affects
Twitch streamers and creators, whose posts, streams and videos are the material in question; Amazon, as the parent company training its AI systems on that material; and, by extension, users of other platforms named in the piece, including Facebook, Instagram and YouTube, where comparable practices have reportedly occurred.
How to use it
From the Twitch website or mobile app, open the account avatar, go to Settings, then Security and Privacy, then Generative AI Training, and switch the toggle off. Note that disabling it only stops that specific use: Twitch says it does not prevent Twitch or Amazon from using a channel's content for other purposes under Twitch's Privacy Notice, such as AI-assisted sponsorship tools, recommendation-based discovery and AutoMod.
How solid is it
The account comes from WIRED, drawing on Twitch executives' own on-the-record remarks during a company livestream, including a named executive's description of his own comment as candid, plus the text of Twitch's current Terms of Service. The reporting is originally from WIRED en Español and appears here in translation from Spanish.
Risks and caveats
The opt-out addresses only the AI-training use case; other data uses continue regardless of the toggle. The article leaves open exactly when the training began, whether Twitch or Amazon's business partners also draw on this content, and to what extent creators' authorship is otherwise respected. It does not name the forum where the 16,000 creators voiced opposition, does not identify which AI models are trained on Twitch content, and gives no total creator-base figure against which to size the 16,000.
“I don't know for sure, but I think it's quite reasonable to assume that almost any publicly available content is used to train models in one way or another, with or without permission. So I think we also need to acknowledge that there's a lot here that's beyond even our direct control.”
— Mike Minton, Twitch's head of product