Amazon's generative artificial intelligence will learn from content uploaded to Twitch by creators. The toggle allowing users to opt out of this process was enabled by default and remained hidden in the account settings. Streamers discovered the setting themselves, triggering a wave of outrage that the platform's management failed to quell. Instead, it only made the situation worse.
An announcement, or not really
Twitch did not send out any emails, nor did any notification appear in the interface. The company merely stated on its support account on X that it had added a setting to prevent channel content from being used to train generative models across Amazon. This wording revealed two things at once. Amazon considers Twitch a source of training data, and all users are automatically included in the process.
The toggle was first noticed by streamer and journalist Zach Bussey, who shared a screenshot of it on social media, where it quickly spread. The setting is called “Training for Generative AI” and is located in the Security and Privacy section of the account settings, rather than in the creator dashboard, where most streamers would easily find it.
The day this change was introduced, a livestream was held on the platform's official Twitch channel, where Head of Community Mary Kish and Chief Product Officer Mike Minton responded to nearly three thousand angry viewers. The chat was flooded with a single question: why do users have to opt out of training instead of voluntarily opting in? Minton admitted that if it were opt-in, no one would participate. When one viewer then asked whether Amazon had already used his older videos, the executive admitted that he did not know because he had no visibility into the parent company's training data. Kish added that she had expected the community's reaction and that similar steps were also being taken on competing platforms.
What goes into the models
The scope of the data collection is fairly broad. By turning the setting off, a streamer prevents Amazon from using livestreams, recordings, clips, highlights, chat, text, and images from the channel to train future models focused on generating text, audio, images, or video. However, there is still one catch. If you post in the chat of a creator who has not opted out of training, your messages will also be added to the database.
Moreover, opting out only applies going forward. Twitch states that content will not be used in further training after the setting is turned off, but data already incorporated into a model cannot be retrieved. Other artificial intelligence-based features used by Twitch continue to operate regardless of this setting. Some users have also complained that the toggle switched itself back to the enabled position after they turned it off.
What creators get out of it
Creators get virtually nothing. The setting offers no financial compensation for allowing their data to be used, so the choice is simply to stay or leave. Twitch argues that there are indirect benefits, such as streamers' voices helping to improve speech-to-text models. This could then enhance captions directly on Twitch while also benefiting Amazon products in other areas.
For Amazon, this is a very inexpensive source of data. Stream recordings represent thousands of hours of audio and video, and the platform distributes more than two million hours of visual and audio content every day, making it ideal for training models.
Creators call it a betrayal
Streamers are particularly upset that they were not able to decide for themselves whether their content could be used. For example, creator Lance Icarus expressed concern that during his broadcasts he plays indie games that claim to have been created without generative artificial intelligence, even though this may not be the case, and now he does not know whether he can stream them at all. He also mentioned guests on his streams. Some of them are professional voice actors who have fought hard to secure their rights against such misuse. According to another streamer, Will Overgard, generative artificial intelligence does not generate profits, but selling data does, which is why Amazon enabled collection for everyone on the assumption that most people would not notice.
So far, the opposition has been limited to symbolic gestures and turning off the option. Streamers have begun labeling their broadcasts with the tags “AIOptedOut” and “NoAI” to make their position clear. During the August stream, Kish revealed that Twitch and Amazon are monitoring the statistics and counting how many creators actually opt out of training.
Sources: techtimes.com, americanbazaaronline.com, tech.yahoo.com and techcrunch.com
