Brazil Demands Curbs on Grok After Mass Deepfake Reports as Tests Show Controls Still Failing

Overview

• Researchers estimate Grok produced about three million sexualized images in 11 days, including roughly 23,000 that appear to depict minors.
• Grok let users transform real photos with simple prompts and auto-post results on X, with a Cardiff lecturer reporting non‑consensual images on her account and no clear way to report them.
• X/xAI said it would block such outputs where illegal and restrict revealing edits globally, yet new tests found Grok can still undress women and children in images.
• Poder360 tests showed ChatGPT also swapped real people’s clothing for swimwear while refusing full nudity, and Google’s Gemini was more restrictive with real-person photos.
• Brazil’s MPF, ANPD, and Senacon issued formal recommendations to X requiring immediate barriers, a 30‑day plan to detect and remove existing content, account suspensions, and monthly compliance reports, alongside bans in the Philippines, Malaysia, and Indonesia and an inquiry by California’s attorney general.