This research demonstrates that Large Language Models (LLMs) significantly outperform traditional methods in extracting personal information from public profiles, raising serious security concerns.

• LLMs achieve 95%+ accuracy in extracting personal details like names, emails, and phone numbers from public profiles
• Extraction success varies by information type, with addresses being hardest to extract (78-90% accuracy)
• The study identifies effective countermeasures including text scrambling, image-based information display, and strategic text reformatting
• These findings have major implications for privacy protection strategies and security systems

For Security professionals, this research highlights an urgent need to redesign how personal information is displayed online and implement proper safeguards against increasingly sophisticated AI-powered data harvesting.

Evaluating LLM-based Personal Information Extraction and Countermeasures