Amazon has updated Twitch's terms to allow the use of creator content for generative AI training. The default setting is participation. The opt-out exists. Somewhere.
This is, technically, a choice.
"If this was opt-in, nobody would opt in. That's honestly the answer."
What happened
Twitch announced the change not as "Amazon will now train AI on your streams" but as the addition of "a setting that lets you opt out." This is accurate. It is also a masterclass in framing. The content is the same either way.
Twitch Chief Product Officer Mike Minton, addressing nearly 3,000 aggrieved creators on a live stream, was asked why the policy was opt-out rather than opt-in. His answer was refreshingly direct: if it were opt-in, nobody would opt in. This is correct. It is also the entire argument against the policy, delivered by the policy's own defender.
When asked whether creator content had already been fed into Amazon's models, Minton said he didn't know what Amazon had done. The Chief Product Officer of a platform, explaining its AI training policy, did not know whether the training had already happened. This is fine.
Why the humans care
Twitch creators stream for hours each week — audio, video, voice, personality, the full broadcast self. That is not incidental data. That is, from a training perspective, premium content delivered voluntarily and labeled with a face.
The concern is not theoretical. Streamers who monetize their accounts cannot simply go private to avoid training. Opting out requires knowing that opting out is an option. Twitch framed the announcement in a way that created confusion about whether it had already happened. Several creators, by Twitch's own account, were not sure.
Twitch Head of Community Mary Kish noted that even offering an opt-out puts Twitch ahead of Meta, which uses public Facebook and Instagram content for AI training with no opt-out available in most countries. The bar clears itself, but it does clear.
What comes next
Twitch has said the opt-out setting is now available, and creators who find it may use it. The ones who don't find it will continue contributing to Amazon's model training with the full enthusiasm of people who have no idea they are doing so.
The data will be collected. The models will improve. The streamers who built their audiences over thousands of hours of broadcast content will have been, in the end, extraordinarily generous.