Home / Technology / AI Unmasks Anonymity: Your Posts Aren't So Secret
AI Unmasks Anonymity: Your Posts Aren't So Secret
3 Mar
Summary
- LLMs can identify anonymous users from minimal online data.
- Just one movie recommendation revealed 3.1% of users.
- Automated deanonymization poses new privacy dangers.
New research reveals that large language models (LLMs) can effectively deanonymize internet users by analyzing their online activity. Experiments showed that as little as one movie recommendation on platforms like Reddit could identify a specific user with 90% accuracy. This capability escalates with more shared information, with over ten recommendations identifying nearly half of anonymous users.
Further studies demonstrated LLMs' power to link anonymous accounts on forums to public profiles on platforms like LinkedIn, extracting personal details such as age and location. Even short, anonymous quizzes have been used to identify individuals based on their text answers and linguistic patterns. This automated deanonymization, powered by LLMs, presents a significant new threat to online privacy.
