Twitch has added a Generative AI Training setting that lets streamers block parent company Amazon from using their streams, clips, and videos to train AI models. The toggle, buried in the Security and Privacy menu, was reported on August 15, 2026, and its arrival confirmed what Twitch had never explicitly disclosed: creator content was already being fed into Amazon's model pipeline by default.
More than 16,000 creators posted opposition in a Twitch forum once the setting appeared, most of them asking the same question — why was this on to begin with. Twitch's Terms of Service, in effect since March 2024, granted the platform and its sublicensees broad rights to use, reproduce, modify, and create derivative works from user content. Those terms did not name AI training as one of those uses.
The opt-out itself is a single click. From the Twitch site or mobile app, a streamer opens their avatar, goes to Settings, then Security and Privacy, and flips Generative AI Training off. The scope of what that toggle actually stops is narrower than it sounds.
Key facts
- 01Twitch added a Generative AI Training toggle under Security and Privacy that lets streamers block Amazon from using their content to train AI models.
- 02More than 16,000 creators posted opposition in a Twitch forum after learning the setting had been on by default.
- 03Twitch's Terms of Service, in effect since March 2024, granted sublicensable rights to user content but did not explicitly mention AI training.
- 04Disabling the toggle does not stop Twitch or Amazon from using channel content for recommendations, sponsorship tools, or AutoMod.
- 05Product head Mike Minton said keeping the default enabled was necessary because otherwise 'no one would participate.'
Twitch notes next to the setting that disabling it does not prevent Twitch and Amazon from using channel content for other purposes described in the Privacy Notice. Those carve-outs include AI-powered recommendation, monetization, and sponsorship tools, plus safety systems like AutoMod. In other words, the opt-out covers generative model training specifically, not the broader machine-learning surface area Twitch already runs on top of creator output.
“no one would participate”— Mike Minton, Twitch head of product
Mary Kish, Twitch's head of community, announced the changes on a livestream and acknowledged the reaction would be negative. Mike Minton, Twitch's head of product, gave what he called a candid response on why the setting shipped enabled by default, saying that otherwise 'no one would participate' in AI training. Minton also argued the practice is not unique to Twitch and that other AI developers are likely scraping the same publicly available material.
Minton went further and named the industry-wide dynamic that most platforms won't put on record. Publicly available content, in his framing, is training data whether or not the platform hosting it has given permission — a candor rare enough from a large platform executive that it becomes the story on its own.
The disclosure gap matters because the Terms of Service update from March 2024 never spelled out AI training as a use case, and Twitch has not said publicly when Amazon actually started ingesting creator content. That leaves streamers looking backward at roughly a year and a half of activity with no clear picture of what was used, by which teams inside Amazon, or whether external partners had access.
The context is a training-data crunch across the industry. OpenAI has signed licensing deals with publishers, including WIRED parent Condé Nast, precisely because open-web scraping is running out of runway. Meta has confirmed using Facebook and Instagram posts to train its models, and has been reported using employee activity and Ray-Ban smart glasses screenshots. Google's use of YouTube content for model training has been documented as well. Twitch's roughly one-and-a-half billion hours of livestream content per quarter is exactly the kind of long-form, conversational, multimodal corpus that frontier labs are short on.
The counterweight is that Twitch has now given creators a working switch, which is more than several peer platforms offer. The opt-out is one click, it is not gated behind an email request or a form, and it exists in the same settings menu as basic privacy controls. Streamers who care about this can act today. What creators cannot do is claw back anything ingested before the toggle existed.
The pattern this cements is the one Amazon, Meta, and Google now share: use first, disclose later, offer an opt-out when the backlash arrives. That is a defensible playbook while training data remains the scarcest input in the model race, but it hardens the case for regulators and courts to treat opt-in consent — not opt-out — as the baseline. The next platform to ship a training-data setting will be judged against a public standard Twitch just helped set, and against 16,000 creators who made clear the default was the problem.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
