Twitch has updated its account settings to let streamers opt out of having their content used to help train the artificial intelligence models of Twitch’s parent company, Amazon.
Although the move has reassured some creators, it remains unclear exactly when the streaming platform began using the posts, streams, and videos of its users to train Amazon’s systems. The revelation is raising new concerns about how big tech companies handle their users’ data.
The opt-out process is simple. From the Twitch website or mobile app, click on your account avatar, select Settings, and then go to the Security and Privacy section in the menu. There, you’ll find the Generative AI Training option, where you can disable the use of your content for that purpose.
In the language next to the toggle, Twitch notes that “disabling this option does not prevent Twitch and Amazon from using your channel’s content for other purposes described in Twitch’s Privacy Notice.” These include AI-powered platform features designed to facilitate streamers’ growth and monetization, such as real-time assistance for sponsorship campaigns, viewer discovery through recommendations, and community safety via tools like AutoMod.
The new setting aims to recognize creators’ right to decide how the content they produce is used. However, its launch has also raised questions about how Twitch and Amazon have used that content up to now.
In a forum dedicated to the topic, more than 16,000 creators expressed opposition to having their content being used by default to train Amazon’s AI systems—a practice that only came to light following the update to the settings.
The backlash arose after a livestream in which Mary Kish, Twitch’s head of community, explained the introduction of the changes. The executive acknowledged that the change would provoke a negative reaction. Mike Minton, Twitch’s head of product, also noted—in what he described as “a candid response”—that keeping the option to use content for AI training enabled by default was necessary, since otherwise “no one would participate” in the process.
Minton pointed out that these mechanisms are not unique to Twitch and considered it likely that other companies developing AI systems are also extracting content from Twitch and other services to use in training their models. “I don’t know for sure,” he said, “but I think it’s quite reasonable to assume that almost any publicly available content is used to train models in one way or another, with or without permission. So I think we also need to acknowledge that there’s a lot here that’s beyond even our direct control.”
These statements raised new questions: Since when has Twitch content been used to train AI models? Is Amazon the only company using this data, or are its business partners also involved? To what extent and in what ways is the authorship of content published by creators respected?
Twitch’s Terms of Service, in effect since March 2024, have stipulated that users grant Twitch and its sublicensees the right to use, reproduce, modify, adapt, distribute, and create derivative works from their content. However, until now, they have not explicitly stated that such materials could be used to train generative AI models.
The Training Data Problem #
This case highlights one of the major challenges facing AI system developers: the growing shortage of high-quality data for training models.
Although companies like OpenAI have reached agreements with various publishers (including WIRED’s corporate parent, Condé Nast) to use some of their content for this purpose, available data sources are dwindling as technology advances and the demand for data increases. As a result, various alternatives have emerged that are reigniting the debate over the ethics and transparency of training processes.
For some time now, Meta has been using posts and images that users share on Facebook and Instagram to train its AI models. Recently, it also came to light that the company was using its employees’ activity and screenshots captured by users of its smart glasses for the same purpose. Similar cases have been documented involving Google and YouTube. Thus, high-quality training data has become another strategic resource—along with chips, memory, and electricity—that the leading developers of artificial intelligence models have found running short as they struggle to sustain the pace of innovation demanded by the market. All signs point to a portion of that demand being met with information generated by users themselves, in exchange for free access to everyday digital services.
This story originally appeared in WIRED en Español and has been translated from Spanish.