What is changing

Twitch, the live streaming platform operating under Amazon, is preparing to use creators' content to train Amazon's generative AI models. Under the new system creators are enrolled automatically; those who do not want their content used must turn the relevant setting off by hand.

The thousands of hours of audio and video that Twitch streams provide are a significant source of training data for Amazon's models. But the setting being on by default has amplified the concern that some creators' content could be used in AI training without their being aware of it.

The company's reasoning

After reactions from the community, Twitch's head of community Mary Kish and product director Mike Minton answered questions in a livestream watched by around 3,000 people.

Explaining why the system was built on default enrolment rather than a model users opt into, Minton said that if participation were opt-in nobody would join the programme. That explanation also shows Twitch is aware of creators' negative attitude toward generative AI.

The new setting also raised questions about whether Amazon had previously used Twitch content in model training. Minton said he did not have information about what content Amazon may have used for training in the past. Kish, for her part, said the use of user content in AI training is not unique to Twitch, recalling that Meta also draws on public content from Facebook and Instagram to train its own models.

How to opt out

Where the setting lives is the practical side of the debate. The steps a Twitch user must follow to leave the system are:

  • Go to account settings rather than the creator dashboard.
  • Move to the Security and Privacy tab.
  • Turn off the Generative AI Training option at the bottom of the page.

That option in account settings permits channel content to be used in training Amazon's generative AI models. Its placement at the bottom of account settings rather than in the creator dashboard that streamers use daily raises the question of how visible the exit really is.

Why it matters

The heart of the debate comes down to a single design choice: default enrolment or opt-in. Minton's reasoning states that choice plainly. The expectation that participation would be low under an opt-in model also contains the assumption that most creators do not want this use. The company is, in other words, predicting which way user preference would go and building the design against that prediction.

The second issue is uncertainty about the past. Introducing a new setting raises the question of what happened before it, and Minton saying he has no information leaves that question open. The practical consequence for creators is straightforward: if they want control over how their content is used, they need to know where the setting is and act on their own.

A third dimension concerns the nature of the content. Unlike edited video or written text, live streaming carries improvised speech, reaction and sustained interaction, which makes it particularly valuable for conversational models. For creators, that same content is often a personal style that constitutes their income. The gap between the data being valuable and its producer sharing in that value is where this debate is most likely to tighten next.