Twitch Turned On Amazon AI Training by Default. Here’s What Creators Can Opt Out Of

Twitch has made creators’ participation in Amazon’s generative-AI training program the default. Unless a creator turns off the Training for Generative AI control, Amazon can use content associated with that creator’s channel for future models designed to generate or synthesize text, audio, images or video. The covered material includes streams, VODs, clips, stream chats, and pictures and text on the channel.

That default—not simply the existence of AI training—triggered the strongest backlash. Creators were expected to find a privacy control and reject participation rather than give affirmative permission first.

What Twitch’s AI-training setting covers

The setting is organized around a channel. When it remains enabled, Twitch says the channel’s content may be used for Amazon’s generative-AI model training, including:

• livestreams and recorded broadcasts;
• VODs, clips and other channel video;
• stream-chat content; and
• images and text displayed on the channel.

This means the setting is broader than a creator’s webcam footage or spoken commentary. Channel descriptions, visual assets and chat attached to the channel can also fall within the defined category of content.

How to turn off Twitch’s Amazon AI training

Creators can disable the feature by opening Twitch account settings and following this path:

Settings → Security and Privacy → Training for Generative AI → turn the toggle off.

On mobile, the equivalent route begins at Profile → Settings → Security and Privacy.

The control is intended to exclude the creator’s channel content from future training of Amazon generative-AI content models. It is not presented as a universal deletion mechanism or as a way to withdraw every form of data processing on Twitch.

Why a creator’s opt-out does not cover every chat message they write

The control follows the channel, not the individual speaker. Turning off the setting on your own account governs content posted to your channel; it does not create an account-wide veto over messages you send in another streamer’s chat.

If you participate in someone else’s stream, that channel’s preference controls whether its chat content is included. As a result, a viewer may opt out of training for their own channel while still appearing in the chat data of a channel whose owner has not disabled the setting.

That channel-based structure is one of the practical limitations creators need to understand: disabling your own toggle does not give you control over material held under someone else’s channel settings.

Mike Minton’s explanation for making the feature opt-out

During a Twitch Patch Notes livestream, chief product officer Mike Minton was asked why the company had made the feature opt-out rather than opt-in. His answer was blunt:

“If this was opt-in, nobody would opt in. That’s honestly the answer.”

The response made the consent dispute more explicit. Twitch’s design assumed participation unless users took action, while Minton’s explanation suggested that an affirmative-consent system would attract little or no participation. The livestream drew nearly 3,000 viewers, and its chat filled with anti-AI messages.

Community head Mary Kish defended the availability of an opt-out, saying Twitch was responding to community members who did not want their material used for generative-AI training. That did not resolve the central criticism from creators: that permission should have been requested before enrollment rather than after it.

Why the Dutch privacy regulator called the practice “very risky”

The Dutch Data Protection Authority urged Twitch users to disable data sharing for Amazon’s AI training and described the approach as “very risky.” The regulator’s stated concern was that AI models could expose personal data to online scammers, who might misuse it.

Livestreams create particular privacy concerns because they can contain identifiable voices, faces, names and details about a person’s surroundings or private life. Once such material enters an AI-training pipeline, people may have limited practical ability to determine where it went or secure its removal from systems built from it. Reporting on the dispute highlighted those concerns alongside the regulator’s warning.

The available reporting does not establish every detail of Amazon’s prior use of Twitch material. Minton also said he could not confirm what Amazon had already used before the new control existed. That uncertainty is part of why the distinction between stopping future training and undoing earlier use matters.

How creators reacted

The default enrollment produced an immediate backlash from Twitch’s creator community. Anti-AI comments appeared during the Patch Notes livestream, and a forum discussion about the policy attracted more than 16,000 participants opposed to default use of their content.

The objection was therefore about consent and transparency as much as model training itself. Creators questioned why the burden of discovering and changing the setting fell on them, particularly when the material can include their voice, likeness, channel identity and interactions with viewers.

What opting out does—and does not—stop

Turning off Training for Generative AI is a targeted restriction. It is meant to keep the creator’s channel content out of future training for Amazon’s generative-AI content models.

It does not necessarily:

• erase content already collected or reverse earlier model training;
• prevent every use of channel content by Twitch or Amazon;
• control chat messages posted in another streamer’s channel; or
• disable all AI-assisted Twitch features.

Twitch’s privacy language continues to allow other uses of channel content described in its Privacy Notice, including features related to sponsorship assistance, recommendations and community safety or moderation tools such as AutoMod.

The practical takeaway is narrow but important: creators who do not want their channel material used for Amazon’s specified generative-AI training should check the setting and turn it off. They should not assume that doing so removes their content from Twitch’s broader data-processing systems—or that it governs everything they say elsewhere on the platform.