Twitch has faced a backlash from users after confirming that content on the Amazon-owned streaming platform is used to train the company’s generative AI models under a setting that is switched on by default.
The platform said on 12 August that users can opt out of the data collection, but the announcement prompted criticism from streamers who questioned why it had been enabled in the first place.
Mike Minton, Twitch’s chief product officer, said the company was “respecting” users by letting them opt out. Asked why the data was collected by default, he said: “If it’s opt-in, nobody would opt-in. That’s the honest answer.”
Minton said during a livestream that users who remain opted in could have any of their channel content used for training, and that the data collected will not be re-sold to other companies.
Twitch’s own FAQs on its use of AI state that unless users opt out, their content could be used to train generative AI models, the type of artificial intelligence that creates new content such as text, images and video.
The company said a person’s audio might be used to “refine models that create speech to text”, which it said would improve automatic subtitles on Twitch streams as well as on Amazon videos.
Some users questioned the implications for game developers, since the setting would also appear to train Amazon’s models on the video games being played by streamers.
Mary Kish, Twitch’s head of community, used the same livestream to show viewers how to disable the training through their channel settings. Users must open the Settings tab in their Streamer Dashboard, click Security and Privacy, and toggle off “training for Generative AI”.
Doing so prevents Amazon from using streams, clips, images, chats and other channel content to train its generative AI models. Some users have since reported that the setting appeared to be switched back on after they disabled it. “Went in and toggled the switch off, exited, went back in, switch is still on,” one said.
Kish said data collection for AI training had become an industry standard. “We don’t expect you to be happy or excited about this. I don’t expect anyone to react to this favourably,” she said.
The livestream’s chat was flooded with hundreds of comments criticising the move. One viewer wrote: “Nobody would opt in because nobody wants to feed AI with our creativity and content.” Another said Twitch had an “opportunity to set an industry standard” by pushing back against data collection for AI. Underneath Twitch Support’s announcement of the feature, one streamer wrote: “On by default is criminal…. the AI narrative push is so draining.”
It is not clear when Amazon began collecting Twitch users’ data to train its AI models. Minton said he did not know whether users’ data had already been scraped for training, adding that he was unsure what Amazon “has done in terms of model training and what they’ve used and not used”.
The use of creative work to train AI models has been contested in the courts. Stability AI won a High Court case brought by Getty Images over copyright claims, while Anthropic agreed a £14m payout to Harry Potter publisher Bloomsbury.
Amazon, which bought Twitch for nearly $1bn (£740m) in 2014, runs a range of AI services and has invested heavily in the technology as it competes with Google, Meta and other technology groups.
