Twitch, the livestreaming platform owned by Amazon since 2014, announced on August 12, 2026 that channel content will be used by default to train Amazon’s generative AI models. Streamers now have an opt-out option, but it is disabled by default — a move that immediately triggered a wave of backlash across the creator community.
🔑 Key takeaways
- Twitch enables Amazon generative AI training on channel content by default
- Livestreams, VODs, clips, chat messages, images and text posts are all covered
- The setting lives under « Security and Privacy », separate from the Creator Dashboard
- Mike Minton says 10 to 25 firms worldwide likely already train on public content
- Amazon already admitted in 2024 to using Twitch data to train AI models
An opt-out toggle that lands badly
The new option, labeled « Allow your channel content to train generative AI content models at Amazon », is automatically enabled for every account. It covers livestreams, videos on demand (VODs), clips, chat messages, images and text published on the channel. To disable it, creators must head to channel settings under the « Security and Privacy » tab and flip the corresponding switch.
According to Twitch, the setting is intentionally kept separate from the Creator Dashboard to maximize visibility. The support page clarifies, however, that opting out does not block AI usage for internal features such as AutoMod, automated captions, or stream summaries — those rely on pre-existing models and fall outside the scope of Amazon’s generative AI training.
A tense livestream with 3,000 viewers
During a broadcast on Twitch’s official channel, community lead Mary Kish and product director Mike Minton fielded questions from roughly 3,000 viewers, whose chat messages overwhelmingly rejected AI training. The conversation quickly turned into a public grilling.
« If this were opt-in, nobody would opt in. That is honestly the answer. »
Mike Minton, Product Director at Twitch
Mary Kish added: « We do not expect you to be happy or excited about this. » Asked about past use of streamer content, Minton said he did not know Amazon’s exact practices: « I don’t really know the answer to that question, because I don’t know what Amazon has done around model training and what they have or have not used. » That uncertainty fueled fears that historical archives may already have been ingested.
Kish explained that the volume of opt-outs would help Twitch demonstrate that the community is not receptive to AI: « The number of opt-outs will help us reiterate that our community is not receptive to AI in any form. »
« Almost every available piece of content » — and 10 to 25 firms
Minton downplayed the impact by pointing to an industry-wide reality: « There are 10 to 25 firms worldwide building cutting-edge language models. It’s reasonable to think that almost every available piece of content is being used to train models, with or without permission. » He added: « If models have been trained on your VODs, it may or may not have been Amazon. »
The product director also pledged that data would not be sold to third parties: « We do not sell your data to other companies that train models. We do not make money from selling your data. »
| Company | Data source | Opt-out available |
|---|---|---|
| Twitch (Amazon) | Livestreams, VODs, clips, chats | Yes, manual in settings |
| Meta (Facebook/Instagram, UK) | Public posts | Yes, manual |
| Meta (Facebook/Instagram, rest of world) | Public posts | Only by setting accounts to private |
Kish reminded viewers that Twitch is not the only company mining public content: Meta uses public Facebook and Instagram posts to train its AI models. She added that Twitch’s opt-out option is « a reflection of this community’s voice saying they do not want their content used to train generative models ».
Earlier admissions and immediate backlash
In 2024, at a conference hosted by The Information, Mike Minton — then director of monetization at Twitch — already confirmed that Amazon was using Twitch content to train AI models: « Yes, certainly, » he said, adding that « within the bounds of user trust and privacy regulations, which vary across the world, we absolutely have a role to play ».
User reactions were immediate. In the livestream chat, many viewers wrote: « Nobody would opt in because nobody wants to feed AI with our creativity and content. » Others argued Twitch had « the chance to set an industry standard » on consent. Kish herself disabled the option on her own channel out of respect for her audience, while Minton admitted that, were he a streamer, he would probably leave it on.
AI innovation vs. creator trust
The policy spotlights growing tensions between how big tech firms use data and the rights of creators. Concerns span consent, intellectual property, data retention and the financial fallout if viewers leave the platform.
Amazon, which is investing heavily in AI to compete with Google and Meta, declined to comment. Twitch’s press email bounced repeatedly when reporters sought clarification. The announcement comes as Twitch remains one of the world’s leading livestreaming platforms, hosting top creators with tens of millions of subscribers.
Conclusion
By defaulting its AI training opt-in on streamer content, Twitch crystallizes a debate that extends well beyond the platform: consent in the era of generative models. Amazon’s move reflects the defensive posture of a player betting on AI to stay competitive against Google and Meta — at the risk of alienating its creator base.
The coming days will show whether community mobilization pushes Twitch to reverse the default setting or tighten its terms. Longer term, regulatory pressure — particularly in Europe with the AI Act — could force the entire sector to rethink opt-in rather than opt-out mechanisms.
Sources
This article is published for informational and educational purposes only. It does not constitute investment advice. Do your own research (DYOR) before making any decision.
