Twitch Is Now Training Amazon's AI on Your Streams by Default — and the Company Knows You Won't Like It

Twitch is now using every channel on its platform to train generative AI models across Amazon, and the only way to stop it is to dig into your settings and switch it off yourself.

The Amazon-owned streaming giant turned the feature on by default, meaning streamers' live broadcasts, saved videos, clips, chat messages, images, and text are all being fed into AI systems that Amazon uses across its business. A help page on Twitch's own site confirms the full scope of what gets collected Twitch Help.

Twitch chief product officer Mike Minton was blunt about why the company made it opt-out rather than opt-in: because if it were opt-in, nobody would opt in, GameSpot reported on 13 August 2026. The admission came alongside a wave of public backlash from streamers and viewers who discovered their content was being used without explicit consent, BBC News and 404 Media reported.

For anyone looking to pull their content out, the process is straightforward but not obvious. Users must open their Twitch security settings and turn off the option labelled "Training for Generative AI," VGC reported. That toggle stops Twitch from using your channel's material for model training.

There are two important catches. First, opting out only covers generative AI training — it does not affect other AI-powered features on the platform. AutoMod, Twitch's automated community-safety tool, and AI-generated captions are classified as platform services, not individual settings, and there is no way to opt out of those, Engadget and GameSpot reported. Second, even if you opt out, any chat messages you post on someone else's channel can still be used for AI training unless that streamer has also opted out, according to GameSpot.

Twitch told users that once they opt out, the company does not retain their data for other types of generative model training beyond what is needed to run the service and its safety features, GameSpot reported.

For context, generative AI models are systems trained on large volumes of text, images, or video to produce new content — writing, pictures, audio — in response to prompts. Companies like Amazon, Google, and OpenAI rely on vast datasets to improve these models, and user-generated platforms such as Twitch represent a rich source of that training material.

What makes this land awkwardly for Twitch is the company's own admission that the choice was strategic. By Minton's logic, the default-on setting exists precisely because voluntary participation would have yielded little data. For streamers who built their audiences on the platform, the idea that their live performances and chat logs are now raw material for Amazon's broader AI ambitions has struck a nerve — and the opt-out toggle, while functional, does not cover every message they have ever typed in someone else's chat.