How to stop Discord training AI on your data

All accounts (free and Nitro)

The captured policy contains no standalone use of "AI", no "artificial intelligence", and no form of "train" at all. Until 1 September 2026 this entry concluded from that the policy "never bridges" content use and model training, and that the question "cannot be determined either way". That went too far, and a re-audit caught it: the document does bridge them once, in the sentence Discord uses to describe moderation. It says it may use "content reported to us, content that violates our Terms of Service, and certain other content widely available on the service (such as public posts, usernames, avatars, banners, user profiles, server names, server icons, and server banners), to create systems and models that can be automated to more swiftly detect, categorize, and take action against prohibited content or conduct". Building models out of user content is what this site asks about, and that is an answer — a narrow one. It covers content you or someone else reported, content that broke the rules, and material already public; reported content can include private messages, which is the part worth knowing. What it does not cover is everything else. For ordinary messages that nobody reports, and for any purpose other than safety enforcement, the document still says nothing, and the clause quoted below — use of content in "larger spaces" to "develop, improve, and power our services" — stops short of saying models are built from it. The other adjacent language is a reference to "automated technologies such as machine learning systems that help us power and improve our services", framed as a description of processing methods under GDPR legal bases. Answered for anti-abuse models, unanswered for the rest.