Twitch Adds a Generative AI Opt-Out Toggle, Enabled by Default

Twitch now lets users opt out of having their content used to train Amazon's generative AI models. The toggle, labeled "Training for Generative AI," appears in account settings under the Security and Privacy tab and was switched on by default when discovered by The Verge on August 12, 2026 The Verge.

According to a Twitch support page, opting out means a user's streams, past broadcasts (VODs), clips, chat logs, channel images, and channel text will not be used in future training of any Amazon AI model whose purpose is to generate or synthesize text, audio, images, or video The Verge. Twitch's own description of the toggle is explicit about its limits, though. Turning it off does not opt the user out of Twitch and Amazon using channel content for other purposes described in the Twitch Privacy Notice. Those other purposes include AI-supported features for streamer growth and monetization, viewer discovery, and community safety tools such as AutoMod The Verge.

AI-supported features that viewers and streamers already rely on, such as captions and safety tools, will continue to function normally for users who opt out. The opt-out is scoped specifically to generative model training, not to the broader suite of machine-learning systems Twitch operates across its platform The Verge.

One detail worth noting for anyone who streams or participates in Twitch chat: when a user participates in chat on another person's stream, that streamer's opt-out preferences govern whether the chat content can be used for training. A viewer who has opted out of generative AI training on their own account is not protected by their own setting while chatting on a channel whose owner has left the toggle enabled. The control is channel-owner-centric, not user-centric, for chat data The Verge.

The default-on position is consistent with how most platforms have approached AI training toggles. Users who want to exclude their content must actively seek out the setting and disable it. Twitch has placed it in the Security and Privacy tab rather than burying it in a developer or experimental features page, which at least makes it discoverable for users who navigate their settings. But the choice to enable training by default means the majority of streamers who never visit that tab will have their content flowing into Amazon's generative AI training pipelines.

The scope of content covered is broad. Streams, VODs, clips, chat logs, channel images, and channel text collectively amount to a substantial multimodal dataset, meaning data that spans text, audio, and video. Twitch's live-chat data in particular is valuable for training models that need to handle informal, fast-moving, and culturally current language. Audio from streams and visual content from clips add further types of data that are expensive to acquire through other means.

The fact that this is Amazon's training pipeline, not Twitch's alone, connects the opt-out to a much larger AI infrastructure. Amazon's generative AI efforts span Bedrock, Titan models, and Alexa-related ML systems. Content from Twitch feeds into that ecosystem. A streamer who leaves the toggle on is contributing training data to Amazon's model development, not just to Twitch-internal features.

The carve-out language in Twitch's toggle description is also worth parsing carefully. The toggle covers generative AI training specifically. It does not restrict the use of channel content for what Twitch describes as AI-supported features for streamer growth, monetization, viewer discovery, and community safety. AutoMod, Twitch's automated moderation tool, falls outside the opt-out. So does any AI-driven recommendation or discovery system that uses streamer content to improve how channels are surfaced to viewers. A streamer who opts out of generative AI training is still fully participating in Twitch's broader ML apparatus for operational and product features. The opt-out narrows one data flow, not the broader data economy of the platform.

The chat-governance model creates an interesting collective-action dynamic. A streamer who opts out protects their own VODs, clips, and channel content from generative AI training. But every viewer in that stream's chat has their messages governed by the streamer's setting, not their own. A viewer who has personally opted out will still have their chat messages on an opted-in channel used for training. This means the effective coverage of the opt-out is fragmented across the platform, with protection varying channel by channel depending on each streamer's individual choice.

Twitch announced the feature through its support account on X Twitch Support and documented it on its help pages Twitch Help.

The toggle is available now in account settings.