Somewhere between your first stream and today, you gave Twitch “an unrestricted, worldwide, irrevocable, fully sub-licenseable, nonexclusive, and royalty-free right” to use, modify and “create derivative works from” everything you broadcast, in any “media channels now known or later developed or discovered.” That text is live on the terms of service page right now. Nobody made you sign anything this week.
What changed this week is that people started arguing about it. On 31 July, streaming reporter @zachbussey posted that “Twitch and Amazon are apparently planning to use streams to train Generative AI,” that multiple sources told him it “will be announced in the coming months,” and that creators “may be able to opt out of some training data.” Apparently. May. Some. Three hedges in three sentences, and no statement from Twitch, Amazon or AWS to go with them.
So you do not know what Twitch is building. You do know what you licensed, and it covers a lot. The argument worth having this week is not about an announcement that has not happened yet. It is about the hundreds of hours of video you have already put on somebody else’s servers, and what, if anything, you can still do with them.
The licence never mentions AI, and that is what makes it strong
Search the Twitch terms of service for “artificial intelligence,” “machine learning” or “generative.” You get nothing. Zero matches, in the document that governs every frame you have ever pushed to their ingest.
That reads like good news for about four seconds. Then you go back to the grant. Twitch does not need an AI clause, because the rights you handed over were written to outlive the words we use for technology. “Create derivative works from” already reaches a model trained on your VOD. “Fully sub-licenseable” already covers Twitch passing that right to somebody else. “Now known or later developed or discovered” covers the thing nobody has named yet. Whoever drafted that paragraph solved this problem years before anyone had to.
The same sentence does something most broadcasters have never noticed. Alongside the content, Twitch takes the “name, identity, likeness, and voice” you submit with it. Your voice, in the terms of the platform you are probably going live on tonight.
There is a real limit in there, and it is worth knowing about. The licence says “nonexclusive.” You keep your copyright and you can license the same footage to somebody else tomorrow. @ashnichrist made that point in her 1 August thread: non-exclusive platform rights leave you with something you can still sell. That is not nothing. It is also a lot less than most streamers assume they have.
Why livestream data is the interesting kind
@ashnichrist’s thread is the clearest read anyone has published on why this rumour is plausible, and she is careful about what it is. “None of this is confirmed, but we can make assumptions,” she writes before speculating about use cases.
Her argument for the value of the data is the part that does not depend on the rumour being true. Livestream archives are multimodal, which means video, audio and text chat arriving together. They are social, which means real people reacting to each other rather than one person talking at a camera. And they are unscripted, which is the same reason she says Reddit archives are worth what they are worth. If you have ever watched your own VOD back and cringed at a forty second tangent, that tangent is the valuable part.
So the incentive is real regardless of what Amazon announces or does not announce. That is a different claim from “Twitch is doing this,” and the difference matters when you decide how to spend your week.
YouTube already turned this into a checkbox
Here is the useful comparison, and it is checkable rather than rumoured. YouTube publishes a help page called Your content & third-party training, and it says plainly: “By default, the third-party training setting is turned off.” The companion page explains the choice you get if you switch it on: “You select from a list of third-party companies to allow or choose to allow all third-party companies.”
Read the title again before you get comfortable. Third-party. The setting governs outside companies, not what YouTube itself does with what you upload. It is a real control with a narrow scope, and it exists today, in a settings page, with a default that does nothing unless you touch it.
That is the thing to hold Twitch to. Not “will you train on us,” which they can answer with a press release. But “where is the toggle, what is its default, and does it reach the videos already on your servers.”
What to do with the streams you already have
@ashnichrist’s practical advice in that thread is two items long and both are cheap. Ask for an opt-in toggle like YouTube’s, publicly, while the question is still open. And keep copies of your own VODs, on the theory that footage you control is footage you can license later.
The second one is the one people skip, because it is boring and it costs storage. It is also the only item on the list that does not depend on a platform doing you a favour. A library you host yourself is a deliberate act everywhere, including with us: on WpStream a recording does not turn into a video on demand by itself, and the broadcaster has to convert it on purpose. That friction is real. It is also the difference between an archive and a folder of files somebody else can re-license.
None of this requires the rumour to be true. The licence is true today, it was true last year, and it will read the same on the morning Twitch either announces something or does not. Go and read the paragraph you agreed to. It takes four minutes, and you will never be more motivated to do it than this week.