Twitch is training AI on livestreams, says it’s an opt out feature as “nobody would opt in”

Twitch has announced it will use livestreamed content to train generative AI models operated by parent company Amazon. The feature is enabled by default, requiring users to manually opt out through buried security settings to prevent their content from being used for AI training. Even chat messages on channels that haven’t opted out will be included in the training data.

The announcement immediately sparked significant backlash from the streaming community. On Twitch’s UserVoice feedback platform, requests to make AI training optional have become the most popular feature request, with streamers and viewers expressing outrage that the feature defaults to enabled rather than requiring explicit consent.

When challenged on the opt-out design, Twitch Chief Product Officer Mike Minton offered candid justification: “If it was opt-in, nobody would opt-in.” Minton and head of community Mary Kish held a livestream to address community concerns, but their explanation that the company chose opt-out because users wouldn’t voluntarily participate only amplified criticism.

The lack of clarity surrounding the scope of data collection has compounded user concerns. While Twitch created an opt-out toggle, uncertainty remains about exactly what user data is being captured and how it will be used in Amazon’s AI systems. The opt-out process requires manual navigation through security settings, which many users may not discover without guidance.

Sources