Twitch, Amazon's livestreaming platform, has announced that streamer content will be used to train the company's generative AI models. The contested part is not that the content is used but how: the setting is on by default, and anyone who wants to be excluded has to act.

What changes

Twitch's updated terms state that user-generated content "may be used for future Gen AI model improvements". Streamers can opt out of this use from their account settings.

As Ars Technica reports, that is where the real story lies: the content had already been used this way for years. What is new is not the use itself but the appearance of an opt-out and the fact that the practice is now written down plainly.

The company's own reasoning

Twitch chief product officer Mike Minton, responding to user feedback on a livestream, was unusually direct when asked why the system was not opt-in:

"If this was opt-in, nobody would opt in. That's honestly the answer."

The sentence puts on record something the industry rarely says out loud. No large platform gathering content for AI training builds an opt-in system, because the data such a system produced would not be anywhere near the scale a model needs. The direction of the default becomes the data policy itself.

What it means for streamers

The practical consequences come in three parts. First, anyone who does nothing is included — and on most platforms the overwhelming majority of users never change a default setting. Second, opting out is not retroactive: there is no visible remedy for content already used over the past years. Third, streamer content is not only speech; a face, a voice, reactions and mannerisms all become data.

The wider context

The move does not stand alone. Large platforms have spent more than a year redirecting user content into their own models, and the argument snags on the same point every time: does the content belong to the user or to the platform? Twitch's terms answer in the platform's favour, but at least leave a door open.

Minton's explanation may end up being the most durable part of the announcement. The reason he gave in defence of the policy is also the case against it: the company already knows most users do not want this.

What streamers can do

Streamers who want out need to turn the setting off in the privacy section of their Twitch account settings. It works at the account level; there is no way to choose stream by stream.

The limits of that choice are worth knowing. Opting out does not undo the fact that past broadcasts may already have been used in training. Twitch's control over clips, recordings and third-party copies is also limited: content downloaded from the platform and moved elsewhere falls outside the reach of the setting entirely.

Where the argument really sits

What stands out here is that the objection turns less on privacy than on copyright and labour. For a streamer, content is not a record but a source of income. A platform using that content to improve its own models means, over time, financing a system capable of competing with the person who produced it.

The concern is not new, but the Twitch case makes it concrete: the distance between the person who creates the content and the party who benefits from it is now the width of a checkbox left ticked.

In brief

  • Scope: user content published on Twitch.
  • Default: on — anyone who does nothing is included.
  • Opt-out: in the privacy section of account settings, at account level.
  • Not retroactive: content already used falls outside the setting.