NEW YORK — What happened: Data protection authorities across the European Union have issued coordinated enforcement notices clarifying that scraping personal information from public websites to train machine learning models violates existing consent frameworks.
Why it matters: The entire generative AI boom relies on massive, unvetted datasets scraped from the web. Restricting this practice threatens the fundamental data diets of upcoming foundation models.
Deep dive: The enforcement focuses on the 'legitimate interest' exemption previously claimed by AI labs. Regulators argue that training commercial models on copyrighted or personally identifiable text without opt-out mechanisms fails the proportionality test under GDPR.
Report check (claims vs what is verified vs still rumor): Tech companies claim they only use aggregated, anonymized data. Privacy advocates and regulatory audits have repeatedly demonstrated that personally identifiable info easily leaks out of models during inference.
Open questions: Will AI labs retreat from the European market, or will they pivot entirely to synthetic data and expensive licensing agreements?