What happened?

Twitch has added a new option in account settings allowing streamers to opt out of having their content used to train Amazon's AI models. Users can click their account avatar, go to Settings, then Security and Privacy, and turn off the 'Generative AI Training' toggle.

Twitch states that turning off this toggle does not prevent the company from using content for other purposes, such as real-time support for sponsorship campaigns, recommendation systems for viewer discovery, or community safety tools like AutoMod. It remains unclear since when content has been used to train Amazon's AI systems.

Why does it matter?

The announcement of the new setting triggered strong backlash from Twitch creators. On a dedicated forum, more than 16,000 streamers objected to their content being used by default to train Amazon's AI.

The controversy grew after Twitch's head of community, Mary Kish, announced the change during a livestream. Kish acknowledged the move would draw a negative reaction. Twitch's head of product, Mike Minton, in what he called a 'candid response,' said keeping the option on by default was necessary, otherwise 'no one would opt in.'

Minton noted that this kind of data use is not unique to Twitch, and that virtually all publicly available content can be assumed to be used for model training in some way, with or without consent. Twitch's terms of service, in effect since March 2024, allow the company and its sublicensees to use, reproduce, and create derivative works from user content, though those terms did not explicitly mention generative AI training.

What about other companies?

This case highlights a core issue facing AI developers: the scarcity of quality training data. While companies like OpenAI have struck deals with some publishers (including WIRED's corporate owner, Condé Nast), data sources are shrinking as technology advances and demand grows.

  • Meta: Uses posts and images shared on Facebook and Instagram for model training; it also emerged that screenshots taken by smart glasses users were used for the same purpose.
  • Google/YouTube: Similar cases of data use have been documented.
  • Amazon/Twitch: Creator content may have been used for training purposes without clear notice to users until the opt-out option was added.

What's next?

Questions remain unanswered about since when the content has been used, whether Amazon's business partners have access to this data, and to what extent creators' copyright is protected. It is unknown whether the company will provide further clarification on these matters.