How To Stop Twitch From Training AI On Your Streams

Twitch recently confirmed it will use streamers’ content to train its generative AI models. The Amazon-owned platform is automatically opting all users in to the feature. And in the most predictable turn of events possible: the move has not gone over well, to put it mildly.

Within hours of Twitch’s announcement, nearly 14,000 people had upvoted a comment on the company’s support forum demanding the platform make the setting opt-in and not opt-out. “We as creators should have the choice whether to use AI or not on our channels,” it said. The post has thousands of comments echoing the sentiment, many of which did so in more forceful terms.

The bit of good news here is that users can opt out of this setting, even though it’s enabled by default. Users can opt out from the web via their account settings by clicking on your avatar (towards the top right of your screen) and heading to Settings > Security and Privacy and scrolling down to “Training for Generative AI.” You can also access the menu directly here.

Likewise, in Twitch’s mobile app, you can find it under Profile > Settings > Security and Privacy. You only need to opt out in one place, so if you’ve already switched it off in the app, you don’t need to do so on desktop and vice versa.

The opt-out applies to training for generative AI models, but not other AI features. As Twitch notes, disabling the setting “does not opt you out of all AI or machine learning uses at Twitch.” The company says it may still use user data for features like AutoMod and captions, though data used for those features is not kept for training GenAI models.

In a Patch Notes stream following the update, Twitch’s Chief Product Officer Mike Minton explained that there are some AI-powered features that are necessary for the platform to function and thus impossible for users to opt out of. “A lot of our systems that make Twitch work now rely on some form of a — usually a specific targeted — AI model, that evaluates text and helps us understand what’s going on across the service,” he said. “For example, nearly every bit of audio is transcribed into text across the entire site. That then enables things like stream summaries to happen. It also helps us from an AutoMod and other perspective.”

According to Minton, opting out won’t impact your reach. “This setting doesn’t… actually impact discoverability,” and it “has zero impact on your ability to be discovered on Twitch or anything like that,” he said. At the same time, he explained, the company uses AI models to improve its understanding of content and how it makes recommendations broadly.

As for the reason why AI training isn’t opt-in, it’s probably exactly what you expect. Speaking on the stream, Minton made it clear that Twitch is well-aware of how unpopular the feature would be. “If it was opt in, nobody would opt in. That’s honestly the answer,” he said. “So it’s going to be on by default. And as I said, almost every service in the world is on by default. I think the thing that we’re doing here that is unique and different is respecting your wishes to opt out of model training. And I know this is not a fan favorite. I know it’s very upsetting to the community, but this is, you know, where we are.”

While frustrated users are already threatening to leave the platform, Minton was quick to point out that people may not have it better elsewhere. Many other platforms rely on users’ content for AI training and don’t offer any kind of opt you.

Google, for example, has confirmed it uses YouTube content to train its AI models, including Gemini and Veo 3. The company told CNBC last year that only a “subset” of its catalog is used for this purpose. But the platform doesn’t make it clear to channel owners when their content is being used in this way and doesn’t offer an opt-out. Meta also trains its models on Facebook and Instagram posts shared by users, including live streams, without an opt-out.

Minton also pointed out that there’s a good chance that other AI labs are scraping and training on Twitch content and there’s little that can be done about it. “I don’t know this conclusively, but I think it’s a pretty reasonable assumption that almost any publicly available content is being used for model training in one form or another, whether permitted or not,” he said. “So I think we also just need to acknowledge that, you know, there’s a lot here that is even beyond our direct control.”