On August 12, 2026, Amazon-owned streaming service Twitch introduced an opt-out setting under Security and Privacy allowing streamers to block the platform from using their broadcasts, clips, VODs, and chat logs to train future generative artificial intelligence models across Amazon. The feature, labeled “Training for Generative AI,” requires users to manually toggle it off.
Navigating the Security and Privacy Menu
The new privacy switch sits near the bottom of the security configuration menu. To disable the feature, users must sign into Twitch, open the account settings, select Security and Privacy, and scroll down to the “Training for Generative AI” control. Disabling this parameter stops Amazon from harvesting channel material—including video streams, highlights, text, and images—for future model development.
According to official statements, the setting applies solely to future model training cycles. Twitch has not disclosed whether Amazon already harvested content from the platform prior to the rollout, nor has the company revealed the specific model architectures involved in the training pipeline.
Distinguishing Operational Tools From Generative AI
Turning off the generative AI training switch does not entirely strip AI infrastructure from a creator’s channel. Twitch continues to process user data for platform operations such as automated captions, recommendation algorithms, monetization tools, and safety systems like AutoMod. According to the platform, these tools execute specific functional tasks without retaining channel content to train foundational generative models that produce new text, audio, images, or video.
Furthermore, individual viewers lack granular control over how their chat messages are handled when participating in live streams. The policy dictates that the channel owner’s setting governs whether messages posted within a specific broadcast can be ingested for training.
Industry Scrutiny and Creator Rights
The introduction of the opt-out mechanism arrives amid a wave of legal battles concerning how technology corporations source copyrighted works for machine learning.
