On August 12, Twitch shifted the default settings on every channel so that a broad range of content (including live streams, VODs, clips, highlights, and the text and images on a channel) can be used to train Amazon’s generative AI models unless a creator opts out. The shift has drawn plenty of attention from the streaming side, and for good reason: creators now have to go find the opt-out setting and actively select it. That conversation is missing a huge piece of the puzzle, though: the developers whose games are the focus of the stream.

Streaming relies on a set of permissions that developers grant, typically through an End User License Agreement (“EULA”) that allows players to record, broadcast and often monetize gameplay footage. Those permissions were written with broadcasting and community building in mind, not with an AI training pipeline sitting on the other end of the platform. Developers who actually own the IP in the streamed games are now facing a major risk that their content is scraped, reused, trained upon, and infringed, without any direct say and without a real means of preventing misuse.

What Changed?

A few mechanics of the new Twitch policy and its impacts are worth walking through.

First, opting out only protects content going forward, so anything already captured before a creator changes the setting remains available for training.

Second, the opt-out doesn’t reach every corner of a channel: chat, for instance, is governed by the settings of the channel where a message was posted, not by the account of the person who typed it. That’s only important to developers in certain instances, for example where an “official” Twitch channel’s chat settings aren’t opted out and the content of the streamed game is being discussed live in detail.

Third, past VODs are not as likely to be used for training as livestreamed content, but they’re not entirely protected. The good news is that Twitch doesn’t save VODs by default, which means a streamer must enable that storage directly. Once they have enabled storage, however, the VODs remain available (for rewatch, download, or potential scraping) for a set period of time after first viewing. That period varies according to account tier: Partners, Prime members and Turbo subscribers retain access to a VOD for 60 days, Affiliates for 14, and everyone else for 7. Functionally that means any new requirements developers put forth to end users and partnered streamers should address those retention windows.

The Risk to IP

Content being streamed on a Twitch channel is almost always someone else’s intellectual property. Any published game contains a vast swath of IP. Not just IP the developer actually created, but IP owned by sound artists, voice talent, and others whose creative work contributed to the end product. On top of that, a developer that’s built a game on behalf of another IP holder is very likely under a contractual obligation not to allow misuse or infringement of the base IP, and so allowing use of the IP through a platform that has instituted an automatic scraping mechanism would be in clear violation of that obligation (whether the “allowing” was direct and intentional or not). Indirect, unintentional allowance might be argued as a minimal shield against claims of misuse and infringement, but that argument isn’t a guarantee of protection.

Twitch’s Ambassador and Influencer programs raise the stakes even further. Running, directing, or paying for a stream that leads to AI training exposes a developer to a much higher degree of risk since the streamed content is produced at the developer’s request and under its supervision or guidance.

What Can Developers Do?

Developers can review their EULA language and add new terms as a prohibitive first step. Those terms should speak directly to AI training or data scraping, whether by disallowing it outright or by requiring streamers to keep opt-out settings enabled.

While updating end-user terms is a good place to start, a blanket prohibition on AI training is difficult to enforce against a broad player base since no developer is positioned to monitor millions of individual streams for compliance. That doesn’t make the new terms pointless. Those terms give developers a basis for requiring compliance for any current or future streaming relationships, especially informal ones that haven’t had their own agreements negotiated directly.

That said, negotiated Ambassador and Influencer agreements offer more security than a EULA and are likely the best practice moving forward for any streaming relationships. Those agreements can be written or amended to require that a partnered streamer has opted out of platform AI training settings before streaming sponsored content, or to build a periodic compliance check into the relationship if it’s ongoing.

Watch This Space…

Streaming platforms will likely keep adjusting default settings as AI training becomes more entwined with their operations, and those changes will happen on the platform’s timeline – not developers’. Building a habit of periodically reviewing streaming permissions in EULAs and negotiated agreements (instead of waiting for the next policy shift to make headlines) puts developers in a better position to respond quickly the next time this happens.

A policy change like this doesn’t call for an overhaul of every agreement a developer has in place, but it’s a good prompt to check whether existing streaming and influencer terms reflect how footage is being used today. Odin Law and Media works with developers on licensing, EULAs, and influencer agreements, and can help advise on how changes like this can impact platform relationships.