Twitch on August 12, 2026 added a setting that lets streamers opt out of having their channel content used to train Amazon's generative AI models — a change that arrives more than two years after Twitch's then chief monetization officer publicly confirmed Amazon was already doing exactly that. Every existing user is opted in by default. The toggle sits in channel Security and Privacy settings at twitch.tv/settings/security, not in the creator dashboard where streamers actually run their businesses.
The scope of what Amazon can ingest is broad. Twitch's updated support page says the covered material includes:
“streams, VODs, clips, stream chats, and pictures and text on your channel [to] be used in future training of a model developed by Amazon whose purpose is to generate or synthesize text, audio, images, or video.”— Twitch, Twitch support page
streams, VODs, clips, stream chats, and pictures and text on your channel, all eligible for use in future training of an Amazon model whose purpose is to generate or synthesize text, audio, images, or video. For a platform where creators routinely broadcast many hours of live audio and video per week, that is one of the largest first-party corpora of long-form human speech and reaction content in existence.
Key facts
- 01Twitch added an opt-out toggle on August 12, 2026, more than two years after an executive confirmed Amazon uses Twitch content for AI training.
- 02Every streamer is opted in by default; the setting lives under channel Security and Privacy, not the creator dashboard.
- 03Covered content includes streams, VODs, clips, stream chats, and channel pictures and text.
- 04Roughly 3,000 users watched Twitch's live Q&A about the change, with chat spammed by anti-AI sentiment.
- 05Amazon has owned Twitch since 2014; Chief Product Officer Mike Minton said he does not know what content Amazon has already used for training.
Amazon has owned Twitch since 2014. The AI-training use was first confirmed at a 2024 event held by The Information, when then chief monetization officer Mike Minton was asked directly whether Amazon trains on Twitch content and answered, "Yeah, for sure." He added that the practice operates "within the bounds of user trust" and applicable privacy regulations. Twitch has not said publicly which years of content have already been used, or which Amazon models were trained on it.
The rollout landed badly with creators. In a livestream on the official Twitch channel, Head of Community Mary Kish and now Chief Product Officer Mike Minton faced an audience of roughly 3,000 users, with chat filled by demands that the setting be opt-in rather than opt-out. Minton addressed the demand head-on.
That is a rare on-the-record admission that a default was chosen because the alternative would collapse the dataset. It is also the clearest statement to date of how large platforms are treating user-generated content in the training-data land grab: as an asset that has to be captured before consent frameworks catch up.
Minton was less clear on what has already been fed in. When one streamer asked whether their past videos had been used, he responded:
That is the Chief Product Officer of the platform saying he does not know what his corporate parent has done with two-plus years of streamer content. Kish framed the new toggle as "a reflection of reacting to this community's voice that you are not wanting to train Gen AI models, and we want to give you that option," and noted Twitch is not alone — Meta uses public Facebook and Instagram content to train its models, with an opt-out available only to users in the U.K.
To turn training off, streamers go to channel settings, open the Security and Privacy tab, scroll to "Training for Generative AI," and toggle it off. Other AI-supported Twitch features — captions, AutoMod, sponsorship-campaign assistance, recommendations — keep running. If a streamer opts out but participates in someone else's chat, the other streamer's setting governs whether that chat message is eligible for training.
The opt-out does not appear to be retroactive. Twitch's language covers "future training," which leaves untouched whatever Amazon has already trained on since 2014, and Twitch has not offered a mechanism to force deletion of a creator's content from existing model weights — a technical near-impossibility, but also a legal question increasingly being litigated by rightsholders in other domains.
The economic frame here is what makes the story matter beyond Twitch. Streaming creators build businesses on their voices, likenesses, and reactions; those same assets, aggregated across the platform, are the raw material for speech-to-text, voice synthesis, and video generation models that could eventually compete with the creators themselves. Twitch offers no revenue share for training use and no per-streamer disclosure of what has been ingested. The opt-out ships the choice to the individual while keeping the default that maximizes Amazon's dataset — a structure that will now be tested against creator backlash, and, likely, against the same copyright and consent lawsuits reshaping the rest of the AI industry.
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.
