Twitch Quietly Enabled Amazon to Train AI
The platform automatically turned on an option that allows posts to be used for developing generative models, and users must turn it off themselves. Executives admit no one would voluntarily agree.
The platform automatically turned on an option that allows posts to be used for developing generative models, and users must turn it off themselves. Executives admit no one would voluntarily agree.
On Thursday, August 13, 2026, Twitch confirmed that it uses content from channels to train Amazon's AI models, sparking a wave of criticism among creators and viewers. The option is enabled by default, meaning broadcasts, clips, images, and chats can be used without prior user consent, according to CNET.
Amazon acquired Twitch in 2014 and is now heavily investing in artificial intelligence to compete with Google and Meta. Material from the platform, especially popular among gamers and younger audiences, represents an extremely valuable resource for developing new models, including projects like Alexa Plus.
Chief Product Officer Mike Minton explained during a stream that users can opt out of data collection, but when asked why Twitch content for AI is not available only with prior consent, he offered a direct answer.
"If it required voluntary opt-in, no one would opt in. That's the honest answer."
Minton added that material from users who do not change their settings may be used for model training, and that the collected data will not be resold to other companies. However, it was unclear when Amazon began taking content from the platform and whether that data has already been integrated into existing systems.
Community Manager Mary Kish showed viewers how to disable content usage through channel settings, stating that data collection for technology development has become an industry standard.
"We don't expect you to be happy or excited about this. I don't expect anyone to react positively."
The stream chat was quickly flooded with hundreds of critical messages. In the opinion of many creators, the platform should ask for explicit permission before using their work.
Users can turn off the option in channel settings, where there is a toggle labeled "Training for Generative AI." Twitch states that even after turning it off, other AI features like Automod and campaign optimization may still be used.
In Twitch's official FAQ, it states: "When you allow training, it means your content may be used for future improvements to generative AI models. An example of what happens when you allow content to be used for training GenAI models is that your audio could help improve models that create speech-to-text, which would improve captions on Twitch, but also captions across Amazon."
Open questions remain about whether user data has already been collected for training. Minton admitted he doesn't know the answer, nor could he explain which materials Amazon has used so far. Disabling the option may stop further collection, but for now it is unknown whether it affects already processed information.
CNET also warns of potential issues with channel chat. Viewers who are not content creators also participate in conversations, and Twitch does not clarify whether their messages are also used for training. This raises questions about audience consent.
An additional problem arises when a streamer re-broadcasts someone else's content. If the original creator has disabled AI training, but the person re-broadcasting their content has the option enabled, the model could still gain access to the original material without the creator's direct consent.
Twitch has faced legal challenges in the past over copyrighted music, and depending on how data is collected, Amazon could also capture copyrighted music from streaming services into its models, creating additional legal issues. BBC has requested a comment from Amazon, and representatives from Amazon and Twitch did not immediately respond to CNET's inquiries.
