The streaming platform Twitch, owned by Amazon, has made a significant change to its policy on the use of user content. A new toggle has appeared in account settings that allows users to prohibit the use of channel materials for training the corporation's generative artificial intelligence models. The key nuance: the option is active by default for all users, which effectively means automatic consent to data collection unless the streamer takes action themselves.

Amazon's training dataset includes not only broadcast archives, but also live streams, clips, text chats, as well as images and text blocks posted on the channel page. According to the service's official help documentation, disabling this setting applies exclusively to future training cycles of models designed to generate text, audio, images, and video. However, the company remains silent on whether training was conducted on this data before the toggle appeared and how much material has already been collected.

When a viewer directly asked which specific data Amazon had used previously, Twitch's Director of Product Mike Minton gave a vague answer, admitting that he did not have that information. This admission only intensified the wave of discontent within the community. The situation escalated to the point that Minton and Head of Community Mary Kish were forced to go live on the platform's official channel, where about 3,000 viewers listened to them.

The most resonant statement was Minton's explanation of the decision-making logic. When asked why the option was not made voluntary (opt-in), he honestly replied: "If it worked on an opt-in basis, no one would opt in." Kish, for her part, tried to present the emergence of the opt-out option as a response to community requests, although it is obvious that this is more of a belated step to mitigate reputational risks.

The audience's reaction was not long in coming: within the first hours after the announcement, a petition demanding that the setting be made voluntary gathered nearly 14,000 votes on Twitch's official feedback platform. This incident fits into the broader context of legal disputes surrounding AI training. Recall that in July, a U.S. court approved a $1.5 billion payment by Anthropic under a class-action lawsuit from authors over the use of their books in training the Claude chatbot.

My analysis: The Twitch situation is a classic example of the "opt-out" strategy, which shifts the burden of responsibility onto the user. Meanwhile, legal precedents such as the Anthropic case show that silent consent to the use of data for AI is no longer a safe zone for corporations. Streamers should immediately check their settings, as their content is a valuable asset that may already be used to train commercial models without any compensation.