Twitch AI Training Setting Is On by Default: How to Opt Out

On August 12, 2026, Twitch added an account setting called Training for Generative AI, and it ships switched on. Twitch AI training is therefore something you leave running by default rather than something you agree to. The switch lives at twitch.tv/settings/security, and Twitch documents it in its own help centre.

Twitch's own wording below was read from its account settings help article and its legal pages in August 2026. Every quote from the help article was re-checked against the live page on September 7, 2026, and none of it had changed. Two quotes from Twitch's chief product officer are attributed to TechCrunch, which reported them.

How to turn off Twitch AI training

1. Go to twitch.tv/settings/security. That is the direct link Twitch gives in its own help article, under "Manage opt-out settings."
2. In the Privacy block on that page, switch Training for Generative AI off.

The setting sits on the Twitch account, not on a channel or a broadcast. If you run more than one account, each one has its own switch.

What the setting actually says

Twitch's description of the toggle, quoted in full from its account settings help article:

Allow your channel content to train generative AI content models at Amazon. Turning this off does not opt you out of Twitch and Amazon using your channel content for other purposes described in the Twitch Privacy Notice, including using AI-supported Twitch features that benefit the community by facilitating streamer growth and monetization (such as real-time sponsorship campaign assistance), viewer discovery (such as recommendations), and community safety (such as AutoMod).

Twitch's help centre also gives an example of what leaving it on means in practice: "your audio might help refine models that create speech to text, which would help improve captions at Twitch but would also help improve captions across Amazon."

What is in scope, and what is not

Twitch answers this in the FAQ attached to the same article. Left on, the eligible content is: "Your stream and the stream's chat, your VODs, your Clips, Highlights, and any text or images on your Channel may be used."

For two of the items in the right-hand column, Twitch explains the distinction as one of retention. On AutoMod: "Twitch will still run AutoMod, which helps keep our community safe, but does not retain your data and use it to produce content." Captions get the same explanation, generating output without keeping your stream "to train a model that can generate new content." Twitch gives no equivalent explanation for recommendations or sponsorship assistance; it simply lists them as unaffected.

Twitch is blunt about why the right-hand column has no switch at all. Under the heading "Why can't I opt-out of all AI uses at Twitch?", it writes: "Some of it runs the service itself, the same way our servers and our code do."

Your chat in other people's channels follows their setting, not yours

Twitch states this plainly: "If you chat on someone else's stream, their opt-out preferences govern if that chat can be used for training."

So the switch is channel-scoped, not person-scoped. Turning yours off does not affect messages you post in other people's channels, because the channel owner's preference governs those. It does cover your viewers' chat in your own stream.

If this matters to your community, the action that changes anything is switching your own channel's setting, not asking viewers to switch theirs.

Opting out applies to future training

Twitch's wording is specific. Opting out means your content "will not be used in future training of a model developed by Amazon whose purpose is to generate or synthesize text, audio, images, or video."

Future training. Twitch publishes no process for withdrawing content from any earlier training, and it does not say whether any took place. Asked directly whether streamers' videos had already been used, Twitch chief product officer Mike Minton said, as quoted by TechCrunch: "I don't actually know the answer to that question because I don't know what Amazon [...] has done in terms of model training."

Minton was more definite about the choice to default the setting on. TechCrunch quotes him from a Twitch Patch Notes broadcast: "If this was opt-in, nobody would opt in. That's honestly the answer."

The practical reading: this switch governs what happens next, and Twitch has not published an account of what happened before.

Twitch updated two legal documents the same day

Twitch keeps public archives of both its Privacy Notice and its Terms of Service, so these changes can be read rather than inferred. Both documents are stamped last modified August 12, 2026, the same day the setting appeared.

The Privacy Notice. The version before it dates to January 15, 2026. Comparing the two, the paragraph on how Twitch uses personal information gained one new passage. It adds that Twitch uses information "to test and improve technology, including machine learning and artificial intelligence, to support and enhance the customer experience across our products and services," and then this sentence:

This includes using data to develop or deploy our generative AI models and services.

The same passage appears in the UK, German and French editions of the notice, each also last modified August 12, 2026.

The Terms of Service. Twitch has not connected anything in this document to the AI setting, and neither will we. The Terms contain no mention of AI, machine learning or model training anywhere, and most of the update covers unrelated ground. Recording it for completeness: the version before it dates to December 8, 2025, and one sentence in the User Content licence changed. It used to end:

Twitch can exercise these rights in connection with monetizing the Twitch Services.

It now ends:

Twitch can exercise these rights in connection with Twitch and/or its affiliates' business.

Keeping the two documents straight matters, because much of the coverage has blurred them. What is documented is that a default-on setting shipped and that the Privacy Notice gained generative AI wording. "Twitch rewrote its terms to take your streams" is a different and stronger claim, and nothing in the Terms supports it.

The notice now points to Amazon's Generative AI Development Disclosure for detail. That page describes Amazon's training data practices in general terms, does not name Twitch, and offers no opt-out of its own. For this particular use of channel content, the Twitch toggle is the control Twitch exposes.

Is it different in the EU or UK?

Twitch documents no regional exception, and we could not find one.

The UK, German and French editions of the Privacy Notice carry the same generative AI passage and the same August 12, 2026 date. Twitch's UK-English help article lists the same Training for Generative AI setting with the same description.

Shared legal wording is not proof that the default is identical in every market, and we are not claiming it is. In every language edition we checked, Twitch describes one setting with one behaviour. If you are in the EU or the UK, open the settings page and check rather than assuming you were treated differently.

It is an account setting, not a software setting

Because the switch sits on the Twitch account, it governs content arriving on that channel regardless of what produced the video. OBS, a hardware encoder, a phone, a cloud platform such as LiveReacting, a pre-recorded playlist: it all lands as channel content, and the same switch applies. There is nothing to change in your streaming software.

Two cases where it is worth checking sooner rather than later. An always-on Twitch channel accumulates broadcast hours, and any VODs, clips and chat that come with them, without anyone logging in to notice. And if you are multistreaming to Twitch and YouTube, this setting covers the Twitch copy of the broadcast only. It says nothing about the YouTube one.

If you are still deciding where your content should live, that is a separate question covered in YouTube vs Twitch vs Kick.

FAQ

Is Twitch AI training on by default? The setting is. Twitch's own help centre describes the control as an opt-out and links to "manage opt-out settings," and TechCrunch and other outlets reported on August 12, 2026 that the toggle appeared already switched on.

Does opting out remove content Amazon already used? Twitch does not say that it does. Its wording covers "future training" only, and it publishes no withdrawal process. Twitch's chief product officer said he did not know whether streamer content had already been used.

Will captions or AutoMod stop working if I opt out? No. Twitch lists AutoMod, captions, recommendations and sponsorship features as unaffected, and explains for AutoMod and captions that they do not retain your content to train content-generating models.

Does the setting cover chat messages I send in other channels? No. Twitch says the channel owner's preference governs chat posted in their stream, so your switch covers chat in your own channel.

Do I need to change anything in OBS or my streaming software? No. It is a Twitch account setting and applies to whatever you send to that channel.

Does this apply to Twitch accounts in the EU and UK? Twitch documents no exception. The UK, German and French editions of its Privacy Notice carry the same wording and the same August 12, 2026 date, and the UK help article lists the same setting.

What to do about it

Open twitch.tv/settings/security, find Training for Generative AI, and make the decision deliberately instead of by default.

It is the same habit that pays off whenever a platform changes something quietly, as it did when Twitch's bot purge dropped view counts. Read what actually changed, then set the switches yourself. For the wider version of that argument, see the risks creators miss behind the camera.