Twitch adds opt-out for Amazon generative AI training
Streamers can now prevent their content from being used to train Amazon's AI models via a new security setting.
Twitch has launched a new account setting that allows streamers to opt out of having their content used to train Amazon's generative AI models. The move formalizes the platform's role in providing massive datasets for the development of AI capable of synthesizing text, audio, images, and video.
The opt-out toggle is located at twitch.tv/settings/security under the Privacy section. By default, the setting is enabled, meaning all users are automatically opted into the training program unless they proactively disable the feature. According to Twitch's FAQ, opting out ensures that streams, VODs, clips, stream chats, and channel-specific text and images will not be used in the future training of models developed by Amazon.
The shift to production training
This policy change follows earlier indications that Twitch was already leveraging its vast repository of human interaction for AI development. In 2024, Twitch Chief Monetization Officer Mike Minton stated that the platform had a "role to play" in training Amazon's generative AI models and noted that content was already being used for AI prototyping. The introduction of the toggle marks a transition from these early prototypes to production-scale training.
Data ownership and consent
The implementation of an opt-out rather than an opt-in system places the burden of privacy on the creators, ensuring Amazon maintains a massive default dataset. This highlights a growing tension between generative AI companies and the creators whose work fuels these models.
Technically, Twitch is drawing a distinction between the "processing" of data and the "retention" of data. The opt-out does not apply to AI-supported service features that process data without retaining it for generative training, such as AutoMod or automated captions, which remain mandatory for all users.
Limitations and outlook
Creators should note that the opt-out is not retroactive. Content that was already incorporated into model development before a user flips the switch remains part of the training set. Additionally, the governance of chat data is tied to the channel owner; if a channel owner has not opted out, the messages of individual chatters may still be used for training even if those specific users have opted out of their own accounts.
As Amazon continues to integrate generative AI across its ecosystem, the industry will be watching whether other streaming or content platforms adopt similar default-in policies or move toward more explicit consent models.