Twitch, the Amazon-owned livestreaming platform, has confirmed that content from its users' channels is being utilized to train Amazon's generative artificial intelligence models. This practice, which has been ongoing for over two years, recently gained attention following the introduction of an opt-out setting, which is enabled by default for all users.
The Dutch data protection authority, Autoriteit Persoonsgegevens (AP), has advised Twitch users to disable this setting, citing concerns over personal data and user control. The AP highlighted that live streams often include personal data such as a streamer's face, voice, name, and images of private spaces, which could be considered sensitive personal data. Once this information is incorporated into Amazon's AI systems, its removal becomes challenging, potentially leading to a loss of user control over their data. User chats and text messages are also subject to this AI training.
Twitch's Chief Product Officer, Mike Minton, acknowledged in an interview that an opt-in model for this feature would likely see minimal user participation, which informed the decision to make it opt-out by default.
The option to disable AI training for channel content can be found in the Streamer Dashboard under "Settings" > "Security and Privacy," then by scrolling down to "Training for Generative AI." Users can toggle off the setting labeled "Allow your channel content to train generative AI content models of Amazon."
It is important to note that disabling this setting only prevents a user's *own channel content* from being used for generative AI training. It does not prevent Twitch and Amazon from using channel content for other purposes outlined in the Twitch Privacy Notice. These other uses include AI-supported Twitch features designed to benefit the community through streamer growth, monetization (e.g., real-time sponsorship campaign assistance), viewer discovery (e.g., recommendations), and community safety (e.g., AutoMod).
Furthermore, a user's decision to opt out does not control how their contributions to *other streamers' chats* are handled. If a user posts in another streamer's chat, whether that chat content is used for AI training depends on that specific streamer's settings, not the individual user's. This means that even with the setting disabled on their own channel, a user's written contributions on Twitch could still become training material if the streamer they are interacting with has the default setting enabled.
While the use of Twitch content for AI training by Amazon has been confirmed by an executive since at least April 2024, the specific opt-out setting was only recently introduced. Prior to this, the use of content for prototyping AI models was acknowledged, though not yet at "production scale."
