Large-Scale Online Deanonymization with LLMs

A research paper demonstrates that LLM agents can identify individuals from anonymous online posts across platforms like Hacker News, Reddit, LinkedIn, and anonymized interview transcripts with high precision. The method scales to tens of thousands of candidates and infers personal details such as location, profession, and interests from a small number of comments. The research was conducted by MATS Research, ETH Zurich, and Anthropic. The findings suggest this deanonymization technique is increasingly practical.