How to Opt Out of AI Training on Meta, LinkedIn, X, and Google

Summary

This practical guide explains step-by-step how EU users can exercise their right to object to AI training on Meta, LinkedIn, X, and Google (GDPR Article 21). It also highlights that an objection only halts future use, as data already fed into models cannot be retrieved.

Why it matters

Most social media users in Europe are unaware that their public posts are being transferred to private AI models under the 'legitimate interest' legal basis rather than explicit consent. The article reveals significant disparities in transparency regarding objection mechanisms: LinkedIn offers a detailed toggle switch, whereas Google provides almost no individual control. Understanding this asymmetry is crucial because the European Data Protection Board’s (EDPB) Opinion 28/2024 explicitly warns that models trained on such data are not automatically anonymized, meaning your data could remain in outputs indefinitely.

Highlights

• The right to object only has a prospective effect. No platform commits to deleting data already used for training. The EDPB notes that models trained on personal data cannot be considered anonymized.
• LinkedIn’s toggle switch only covers comments and ratings in the platform’s own AI features and non-generative machine learning models. A full objection requires filling out a separate GDPR objection form.
• Google does not offer an individual objection option for public web data, providing only a token (Google-Extended) to be added to robots.txt files for site owners. X buries AI training clauses in a single paragraph of its privacy policy, with no named control mechanism for Grok.

Critical take

While the article is well-sourced, it ultimately exposes an uncomfortable reality: the 'object' framework is not a meaningful user right but merely a compliance checkbox. The EDPB’s own Opinion 28/2024 demands unconditional exclusion from the outset, yet none of the four platforms meet this standard. The author’s call to 'object now' attributes too much agency to users. Even a timely objection changes nothing regarding content processed into models over months or years. The outputs of these models remain opaque and unauditable.