Streamers who have spent years building their channels — the VODs, the late-night marathons, the chat chaos, the clips the community clipped and re-clipped — may be surprised to learn that archive could reportedly be feeding Amazon’s AI pipeline. Twitch has added a setting called “Training for Generative AI“ that lets you opt out of having your channel content used to train Amazon’s generative AI models. The setting is live now. According to multiple reports, the practice it addresses has been underway for some time.
Your Content Has Already Been Working Overtime
Here is exactly what the new toggle covers, what it does not, and where to find it.
Find the setting under Settings > Security and Privacy on desktop, or Profile > Settings > Security and Privacy on the mobile app. It covers streams, VODs, clips, chat logs, pictures, and channel text — essentially everything published to your channel. What it does not cover is worth noting: AutoMod, captions, and recommendation features continue using your data regardless of how you configure this toggle. Those AI tools keep running no matter what you select.
Key mechanics every streamer should know:
- The setting is opt-out, not opt-in — you must actively change it or your content is included by default
- Covered content includes streams, VODs, clips, chat, images, and all channel text
- If you chat in someone else’s stream, that creator’s opt-out preference governs your chat — not yours
- Opting out applies to future training; Twitch has not publicly addressed historical use
- AutoMod, captions, and recommendation features continue using data regardless of your preference
“If it was opt in, nobody would opt in.”
The Admission That Came Before the Setting
A candid moment at a 2024 industry event revealed more than Twitch’s official communications ever did.
At a 2024 Information event, Twitch executive Mike Minton reportedly acknowledged that Amazon uses Twitch content to train AI models, saying, “Yeah, for sure” — then said he did not know specifically what Amazon had done in terms of model training. Twitch community head Mary Kish later framed the new toggle as a direct response to creator feedback and concern about generative AI training. It is the Terms of Service scroll problem made tangible: you may have agreed to something nobody clearly explained, except this time an executive said part of it out loud at a panel.
Twitch has not disclosed how long the training practice has been active, which Amazon models are involved, or what historical content was already used.
Go change your setting now — Security and Privacy, on desktop or mobile. The bigger question the opt-out does not answer is what Amazon already holds: which models benefited, which content was absorbed, and what those years of streaming may have already built inside Amazon’s AI infrastructure.
