Streaming platform Twitch is now using creators’ content to train Amazon’s generative AI models by default. The change, which sparked swift community backlash, forces users who wish to protect their streams, clips, and chats to manually toggle off a hard-to-find setting in their account controls.
The quiet rollout caught creators off guard after streaming reporter Zach Bussey posted a screenshot of the newly added option on August 12. The discovery triggered intense criticism across social channels and a flood of negative comments during an official livestream hosted by Twitch executives. Viewers questioned why the data collection was enabled automatically rather than requiring affirmative permission.
“If it’s opt-in, nobody would opt-in. That’s the honest answer.”
Photo: arstechnica.com
Mike Minton, Chief Product Officer at Twitch, via BBC
During the stream, Mike Minton confirmed that Amazon had been utilizing Twitch data for AI training for more than two years. When pressed about whether user content had already been scraped, Minton stated that he did not know what Amazon had utilized in past model training (adding that he was unsure what Amazon has done in terms of model training). Mary Kish, head of community at Twitch, acknowledged the widespread frustration but defended the platform’s reliance on certain automated technologies, noting that core safety systems run the service itself, the same way our servers and our code do (per Twitch’s documentation).
The new privacy control is specific in its application. Toggling the switch off prevents future generative AI content models from incorporating a channel’s streams, video-on-demand files, clips, and chat messages into upcoming training cycles. However, the documentation clarifies that the protection does not extend to chat messages posted in other channels that have not opted out.
Photo: bbc.co.uk
Furthermore, turning off the setting does not stop Amazon and Twitch from using channel content for other operations governed by the privacy notice. Features designed for streamer growth, recommendations, and moderation tools like AutoMod continue to process data because they do not retain information to train generative models (such as features that process information to deliver a service). Twitch published examples explaining that audio data might help refine speech-to-text models to improve automated captions across both Twitch and Amazon ecosystems.
For creators seeking to disable the feature, Twitch places the toggle under the security and privacy tab within general channel settings rather than the main creator dashboard.
How To "Decline" Twitch's Use of Your Likeness As A Blueprint for Amazon's New Gen AI Model.
