Twitch users can now block their streams, VODs, clips, and chat from being used to train Amazon’s generative AI models, under a new setting the Amazon-owned platform added to its account controls. The toggle, labeled “Training for Generative AI,” sits in the Security and Privacy tab of account settings and governs whether channel content feeds future models built by Amazon to generate or synthesize text, audio, images, or video, according to Twitch’s account settings documentation.
The control appears in Twitch’s account settings documentation, which spells out what the setting does — and what it deliberately does not. What the setting actually does (and what it deliberately does not do) is spelled out in the documentation, and the limits are as consequential as the opt-out itself.
What the Opt-Out Covers
Switching the toggle off covers a defined list of material: a user’s streams, VODs, clips, stream chats, and the pictures and text on their channel. Any of that content is then excluded from future training of Amazon models whose purpose is generating content. The FAQ Twitch published alongside the setting gives a concrete example of what leaving it on permits: a streamer’s audio could help refine models that create speech-to-text, improving captions on Twitch while also improving captions “across Amazon.”
Chat follows the channel owner’s choice, not the chatter’s. If a user participates in someone else’s stream, the documentation states that the channel owner’s opt-out preference governs whether that chat can be used for training. An individual who opts out but chats on a channel that has not opted out has no separate protection for those messages.
The control is also not retroactive in its wording. The FAQ frames the exclusion around “future training,” leaving untouched any content already incorporated into model development before a user flips the switch.
What Continues Whether Users Opt Out or Not
The setting stops at generative models. Twitch’s own description of the toggle, quoted in its settings documentation, states that turning it off “does not opt you out of Twitch and Amazon using your channel content for other purposes described in the Twitch Privacy Notice,” naming three categories of AI-supported features that continue regardless: streamer growth and monetization tools such as real-time sponsorship campaign assistance, viewer discovery systems such as recommendations, and community safety systems such as AutoMod.
The FAQ draws the line at retention and generation. Features like AutoMod and automated captions process content to deliver a service but, per Twitch, do not retain that content to train models capable of producing new videos, images, or text. Users get individual controls over some of those features (streamers can adjust AutoMod and captions settings separately), but others have no per-account switch at all.
Asked why users cannot opt out of every AI use on the platform, the FAQ answers that some AI “runs the service itself, the same way our servers and our code do,” and that turning off safety and security systems for one account “would weaken safety and security for everyone else on the platform.”
The Consent Question the Toggle Leaves Open
The setting arrives with the default doing the heavy lifting. The toggle grants permission when left on, and the documentation describes opting out as the action a user takes, meaning accounts that never visit the Security and Privacy tab are, by the design of the control, permitting their content to train Amazon’s models. Under Twitch’s design, opting out is the action a user takes, so accounts that never visit the Security and Privacy tab leave the training permission in place.
That design sits on top of a data-sharing structure that predates it. Twitch’s Privacy Notice, last modified January 15, 2026, states that personal information collected by Twitch “may be combined with personal information about Amazon customers” and used by both companies “to operate, provide, develop, and improve Amazon’s and Twitch’s products and services.” The new toggle carves one use — generative model training — out of that broader flow, while leaving the rest of it intact. The notice also commits Twitch not to materially weaken protections for previously collected data without affected users’ consent, a clause that will matter if the training program expands.
The mechanics draw a distinction Twitch’s own FAQ states: AI-supported features ‘do not retain your content and then use it to train models that create new videos, images, or text.’ Twitch’s documentation adopts that distinction as its organizing principle: features that process are mandatory, training that retains is optional, while the EU AI Act’s transparency rules, examined in What the EU AI Act Doesn’t Tell Founders About Training Data, leave retention and training outside their scope. Related enterprise-control questions around AI infrastructure are covered in Claude Code Sessions Can Now Run on Infrastructure Your Team Controls, where Anthropic’s self-hosted beta keeps session data on customer machines but requires inference to stay with Anthropic.
Where the Control Lives and What Happens Next
The toggle is managed at twitch.tv/settings/security, per the help documentation, under the Privacy section alongside existing controls for whispers, gifts, and badge visibility. No separate announcement post accompanied the change; the documentation itself is the record of the policy, and the setting is live for users now.
What remains unwritten is enforcement and verification. The documentation does not describe an audit mechanism, a confirmation to users that their content was excluded from a given training run, or a process for the retroactive question: content already used before August 12, 2026. What Twitch has committed to in writing is narrower: a defined scope of content, a defined category of models, and a defined set of carve-outs where the opt-out does not reach. For streamers deciding whether to flip the switch, those written boundaries are the policy.
