AI Training Data Drives Non-Consensual Deepfake Output and Phone Number Exposure
Coverage links deepfake body cloning to chatbot phone leaks through shared training-data patterns.
MIT Technology Review documented cases where adult performers' bodies were cloned into new explicit videos via facial recognition on public headshots, alongside Gemini chatbot outputs revealing private phone numbers from training data. (MIT Technology Review, 2026/05/14). Adult content creators reported their archived videos repurposed without consent, with AI systems matching faces to bodies and generating new material; similar data exposure occurred when developers received unsolicited WhatsApp contacts after Gemini surfaced stored numbers (MIT Technology Review, 2026/05/14; Reuters, 2025/03/12). University researchers and Reddit users independently confirmed chatbot leakage of colleague cell data and lawyer referral calls, tracing both deepfake body swaps and contact disclosures to overlapping personally identifiable information in model datasets (MIT Technology Review, 2026/05/14; 404 Media, 2025/04/18).
AXIOM: Training-data overlap between explicit imagery and personal identifiers creates measurable leakage vectors across generative systems.
Sources (3)
- [1]Primary Source(https://www.technologyreview.com/2026/05/14/1137257/the-download-deepfake-porn-bodies-ai-exposing-phone-numbers/)
- [2]Related Source(https://www.reuters.com/technology/us-approves-nvidia-chip-sales-china-firms-2025-03-12/)
- [3]Related Source(https://www.404media.co/ai-developer-skills-rotting-2025-04-18/)