Twitch and Amazon Reportedly Plan to Train AI on Creator Streams. For Now, It Is a Leak, Not an Announcement

Streaming-industry journalist Zach Bussey wrote on July 31 that, according to multiple sources, Twitch and Amazon are preparing to announce generative AI training on creator streams, despite internal friction within the company. Creators would reportedly get a partial opt-out, but not from the full training dataset. Neither Twitch, Amazon, nor Amazon Web Services has officially confirmed or denied this.

What Bussey actually said

Bussey regularly covers Twitch's decisions firsthand, including earlier changes to the platform's ad policy. In his post, he also cited context from two years earlier: in 2024, Twitch's Chief Product Officer, Mikey Minty, said outright at The Information's Creator Economy Summit that Twitch has "a role to play" in training Amazon's AI. Around the same time, Twitch CEO Dan Clancy repeatedly said publicly that he wants generative AI on the platform, describing it as opening up "endless possibilities."

Why this specific report struck a nerve

Those "role to play" statements from two years ago sounded vague enough to easily dismiss as corporate language without any real substance. Bussey's report is different because it talks about a specific, planned announcement, with a real timeframe measured in months rather than an indefinite future. That shifts the story from "the company is considering AI" to "the company is preparing to announce AI," which matters a great deal to creators even without a single line of official confirmation.

Twitch's current terms of service never use the words "artificial intelligence" or "machine learning," but that doesn't mean the topic isn't already covered. By publishing a stream, a creator grants Twitch a license to "create derivative works" from it and to sublicense that right to third parties, plus a separate grant covering their own name, likeness, and voice. That kind of broad, technology-neutral wording is enough to cover AI training without adding a single new clause, and that's exactly the problem: creators today have no clear, explicitly named reference point to check exactly what they agreed to.

How other platforms handle the same issue

Since December 2024, YouTube has given creators a dedicated setting in YouTube Studio to knowingly allow specific third-party companies to train AI models on their content. That consent is off by default, so without manually turning it on, a creator is automatically excluded from the training set. The model Bussey describes would work the opposite way: creators would reportedly be included by default, with opt-out only covering a selected "portion" of the data, not all of it.

Community reaction: from resignation to calls for regulation

Bussey's post racked up several million views and hundreds of comments within days, in wildly different tones. Some creators responded with resignation or dark humor, some openly said they don't mind. Other, more strategic voices showed up too: VTuber Rosa Nagashi advised the community against a mass exodus to other platforms, arguing that switching wouldn't solve anything, since similar training practices likely apply to competitors as well, including Kick, and that a real solution would look more like legislation covering every major streaming platform at once, rather than fleeing from one platform to another that's quietly doing the same thing.

In the same thread, another industry commentator, Ashni, announced she was digging into similar practices on Kick, suggesting the issue might extend across the whole streaming industry rather than being unique to Twitch and Amazon.

What this means for the average streamer

If the leak holds up, the phrase "AI trains on streams" doesn't say much on its own. What matters more is exactly what data, what scope of opt-out, and what impact on a creator's image. Training generative AI on live video and audio could, in theory, capture a creator's voice, way of speaking, facial expressions, and characteristic mannerisms, exactly the elements that build their recognizable personal brand. A partial rather than full opt-out, as Bussey describes it, would mean creators don't have full control over which pieces of their work end up in the training set.

For the average streamer, this is a story to watch for now: no platform has announced the exact scope of training or the final shape of any opt-out yet. The real impact on any specific account will only show up alongside an official announcement spelling out exactly what goes into the training set, and under what terms.

SPAWNSY verdict