이미지: TechCrunch AI
Summary
- Twitch will use creator content to train parent company Amazon's generative AI models, with training enabled by default and creators required to opt out if they don't want to participate
- Twitch Chief Product Officer Mike Minton, asked on a live stream why the company didn't go with opt-in instead, said "if this were opt-in, nobody would do it"
- Meta also uses public posts for AI training, but only UK users can opt out separately — users elsewhere have no option but to make their accounts private
- 정책 방향
- 트위치 콘텐츠, 아마존 생성형 AI 학습에 기본 허용
- 거부 방식
- 크리에이터가 설정에서 직접 옵트아웃해야 함
- 발언자
- 트위치 CPO 마이크 민튼, 커뮤니티총괄 메리 키시
- 발표 방식
- 약 3,000명이 지켜본 공식 트위치 채널 라이브 방송
- 유사 사례
- 메타는 공개 게시물 학습, 영국 이용자만 옵트아웃 가능
What happened
Twitch has decided to use creators' stream video and audio to train parent company Amazon's generative AI models. The problem lies in how it's being rolled out. Rather than framing this as a new feature, the company announced it as having "added a setting that lets you opt out of having your channel content used to train generative AI models." In other words, training is enabled by default, and creators who don't want to participate have to actively turn it off themselves.
About 3,000 people flooded a live stream on Twitch's official channel with chat messages protesting the move. When asked why the company didn't go with opt-in, Twitch Chief Product Officer Mike Minton didn't dodge the question. "If this were opt-in, nobody would do it," he said. When someone asked whether their videos had already been used for training, Minton replied, "I don't even know what Amazon has used to train which models."
What this means
Twitch streamers broadcast their faces, voices, and gameplay footage live for dozens of hours a week. It's a platform where dense audio-visual data accumulates daily, more so than almost anywhere else. For Amazon, this data is a massive supply of real-world content for training its AI models. But for creators, it also means their broadcast assets could be funneled into the company's AI business without their consent.
The generative AI industry has a long history of scraping books, images, and videos from the internet for training without consent. The Twitch community has generally been critical of this practice, and Twitch is aware of it. Community lead Mary Kish acknowledged that the opt-out setting itself was "a response to this community's voice." At the same time, she noted this isn't a Twitch-only problem. Meta already uses public posts from Facebook and Instagram for AI training, and only UK residents can opt out separately. Users elsewhere have no option but to make their accounts private — which isn't really a choice for creators who need public accounts to earn income.
What changes now
Twitch creators need to actively find and enable the AI training opt-out option in their settings menu. Doing nothing means automatic inclusion in training data. Looking at the bigger picture, major platforms are increasingly shifting from "asking for consent" to "granting the right to refuse" when folding user content into AI training resources. This runs counter to Snapchat's move to stop rewarding fully AI-generated content. One trend tries to filter out AI-made output, while the other tries to pull human-made original content into AI training. For creators, both trends demand attention.
