Udemy forces creators to use opt-out windows, the first of which was three weeks and ended on September 12, to prevent their videos from being used to train AI
https://www.404media.co/... X: Jason Koebler / @jason_koebler : This is a really wild story in part because Udemy is being more transparent than other companies, and is saying: It is too expensive for us to delete your data once it's already been scraped. So what are the other companies doing? [image] Ed Newton-Rex / @ednewtonrex : Companies that care about their users' views on AI training give them an opt-in. Opt-outs are designed to be missed. If you genuinely believe your AI training is good for the users whose work you're training on - and you believe they actually read emails about your AI training Ben Williamson / @benpatrickwill : Can only imagine more examples following fast of edtech platforms capitalizing on users' content for generative AI training - followed by a flood of even more edu-bot features and synthetic learning content
Context & Ripple Effects
Udemy’s approach sits in an emerging split between permission mechanisms and commercial access to training material. Google had already offered publishers a robots.txt control for AI training while preserving search visibility, while Stack Overflow signaled that high-value community data could become a paid input for model developers.
The key distinction is timing: Udemy ties creator control to a limited pre-use window and says deletion becomes costly after scraping. Later, YouTube’s creator authorization tool for third-party AI training illustrates a more affirmative permission model.
First-order effects
- Udemy creators must monitor and act within designated windows to keep course videos out of the platform’s AI-training use; those who miss a window face a materially weaker practical remedy once material has been scraped.
- Udemy gains a standardized process for collecting training permissions, but its stated deletion constraint makes that process consequential before data enters its training pipeline.
Second-order effects
- The policy makes consent design a competitive issue for course and creator platforms: an opt-out deadline is less creator-protective than a durable opt-in or authorization control, and creators can compare platforms on that basis.
- It also sharpens the value of provenance and deletion workflows for AI-data suppliers, because a dataset that cannot readily honor later exclusions creates platform and creator-relations risk.
Third-order effects
- If limited opt-out windows become common, control over creative work may increasingly depend on platform-set notice and timing rules rather than continuing creator consent.
- The contrasting approaches in related coverage—publisher controls, paid data access, and explicit video authorization—point toward a fragmented market for training rights, where permission mechanisms become part of platform governance.
The trend: AI-training rights are shifting from an assumed byproduct of hosting content toward a platform-level choice over consent, access, and the reversibility of data use.