Preventing Twitch from Using Your Streams for AI Training: A Guide

Twitch has officially announced that it will utilize streamers’ content to enhance its generative AI models. The Amazon-acquired platform has decided to automatically enroll all users in this feature, a decision that has been met with significant backlash.

Shortly after the announcement, around 14,000 users expressed their discontent in a support forum comment advocating for an opt-in option. The message emphasized that creators should have the freedom to decide whether to allow AI use on their channels, a sentiment that spurred numerous comments in agreement, many emphasizing their frustration more vehemently.

The good news is that users can opt out of this feature, even though it is set to ‘on’ by default. To withdraw consent via a web browser, users should click their avatar at the top right, navigate to Settings > Security and Privacy, and scroll to “Training for Generative AI.” This menu can also be accessed directly in the settings.

In the mobile app, users can find the same options under Profile > Settings > Security and Privacy. Opting out at one location is sufficient, meaning if users disable it in the app, they don’t have to do so again on desktop, and vice versa.

Choosing to opt out pertains specifically to training for generative AI models and does not encompass all AI functionalities. Twitch clarifies that disabling this setting “does not exempt you from all AI or machine learning applications at Twitch.” User data may still be utilized for features like AutoMod and captions, but the data used for these functions will not contribute to training generative AI models.

In a discussion following the update, Twitch’s Chief Product Officer, Mike Minton, explained that various necessary AI-driven features integral to the platform cannot be opted out of by users. He mentioned that many systems in place at Twitch now depend on specific AI models that analyze text, enhancing the platform’s operation. For example, virtually every audio segment is transcribed into text, allowing features like stream summaries and AutoMod functionalities.

According to Minton, opting out will not affect user reach. He stated that the setting “doesn’t… actually impact discoverability and has zero effect on your visibility on Twitch.” However, he mentioned that AI models are employed to enhance content understanding and broaden recommendations.

Regarding the choice to make AI training a mandatory feature, Minton indicated that the decision stems from an understanding of user reactions: “If it was optional, no one would agree to it. That’s the reality.” He noted that it will remain on by default, as many platforms follow a similar approach, emphasizing Twitch’s unique stance of allowing users to withdraw consent for model training, albeit acknowledging the community’s displeasure.

While some users are contemplating leaving the platform in light of their frustration, Minton highlighted that alternatives may not provide better options. Numerous other platforms also utilize user content for AI training without offering any opt-out possibilities.

For instance, Google has confirmed that it employs YouTube content for training its AI models, including Gemini and Veo 3, although it only uses a portion of its catalog. However, YouTube does not alert channel owners when their content is utilized in this way, nor does it offer an opt-out provision. Similarly, Meta leverages user-generated content on Facebook and Instagram, including live streams, for training without providing an opt-out option.

Minton also noted the likelihood that other AI developers are scraping and using Twitch content for training purposes, highlighting a point of concern. “While I can’t assert this definitively, it’s reasonable to assume that virtually any publicly available content is being used for model training in some capacity, whether authorized or not,” he shared. “Thus, we must accept that a lot of this is beyond our direct control.”