Twitch introduced a setting that by default allows Amazon to use streamers' channel content to train generative artificial intelligence models. The setting is enabled automatically, and thousands of hours of audio and video data are already available for training until streamers manually turn it off.
What happened
Twitch enabled the use of streamers' content by default to train Amazon's generative AI models. Streamers must independently go to their channel settings, open the Security and Privacy tab, and turn off the Training for Generative AI toggle. Twitch CPO Mike Minton justified the decision on a live stream in front of three thousand viewers with the phrase: 'If it were opt-in, nobody would turn it on.' The setting applies globally to all channels — both new and existing.
Context
A similar opt-out model is already used by Meta on Facebook and Instagram, where user content is used by default to train AI. The opt-out model instead of opt-in has become an industry trend: platforms argue that voluntary consent yields zero participation. Amazon, as the owner of Twitch, gains immediate access to a massive array of multimodal data — video, audio, and text from millions of streamers — without additional costs for collecting datasets.
Why this matters for the industry
Amazon instantly strengthens its position in training multimodal models, gaining a unique volume of real audio-video content unavailable to competitors Google, OpenAI, and Anthropic. For ML engineers using Amazon models in production, a risk emerges: base models may be trained on content with an unclear legal status, creating a potential vulnerability during commercial deployment. It is expected that next-generation Amazon models will include streaming content in their training data, which could improve capabilities in speech, video understanding, and real-time generation.
Why this matters for users
If you stream on Twitch, your content is already being used to train Amazon's AI. To turn it off: channel settings → Security and Privacy tab → turn off the Training for Generative AI toggle. Important: turning off this setting does not cancel other forms of data use by Amazon according to Twitch's privacy policy. Individual streamers whose voices and images have already been included in the training data could potentially be reproduced in voice cloning and content generation models.
What is still unknown / limitations
Amazon does not disclose technical details: which specific models are trained on Twitch data, what architecture they use, and exactly how streaming content is integrated into the training pipeline. Stream content is not an annotated dataset, but a noisy, unstructured flow, which raises questions about the share of useful signal after cleaning and filtering. The legal consequences of the opt-out option for AI training on user content have not yet been determined by case law.
Sources
- IGN — Twitch Now Trains Amazon's Generative AI Models on Your Channel by Default
- TechCrunch — Amazon will train on Twitch streamers' content by default
Author
Look at AI, editorial team