Twitch Adds Opt-Out for Amazon AI Training Use

Twitch has introduced a new control that lets creators opt out of having their content used to train Amazon’s generative artificial intelligence models. The mechanism covers multiple forms of activity on the platform, including live streams, video-on-demand uploads, clips, and chat logs.

The opt-out is not presented as an onboarding prompt. Users must manually navigate to Twitch security settings to restrict the platform’s use of their data for model development.

What the opt-out covers, and how users must enable it According to the policy description, the scope of the setting includes both media and text generated on Twitch. That means audiovisual content from broadcasts and saved videos, short-form clips, and written interactions via chat are all within the data types referenced.

Twitch’s approach requires individual action by users to change the setting. The company’s configuration is currently set so users are enrolled by default in the relevant data-processing program.

Default enrollment and questions around disclosure

The default participation model is not newly introduced, according to the source material. It says this practice has reportedly been in effect for several years, implying that content may have been included in training processes without users first making an explicit choice to participate.

The retroactive nature of this arrangement, combined with a default opt-in posture, has raised concerns about transparency and user consent. The source notes that the duration and full scope of the training program were not previously disclosed to users in a clear way.

Twitch links usage to broader Amazon generative AI capabilities Twitch has said the data use is intended to improve Amazon’s generative AI capabilities. The company specifically tied this to functions that include speech-to-text and captioning services, and it framed the benefit as extending across the broader Amazon ecosystem.

These use cases matter for creators and viewers because accessibility and transcription features can influence how content is consumed and discovered. At the same time, the policy shift underscores that improvements to platform tools can rely on underlying data practices that users may not notice unless controls are clearly presented and explained.

Governance and reputational risks still in focus

The new opt-out provides a clearer path for creators who want to restrict training use. However, the source material indicates that earlier limited disclosure around how long the program has operated and what it covered may continue to create governance and reputational risks for the platform.

How widely creators adopt the opt-out, and how Twitch communicates the setting over time, remains an open question based on the information provided. The change nonetheless marks a formal acknowledgement that creator data and content are part of generative AI development, and that consent controls are now part of the product experience.