LLMs can unmask pseudonymous users at scale with surprising accuracy

Researchers have found that AI can be used to unmask pseudonymous users on social media platforms with high accuracy, potentially compromising online privacy. The success rate of identifying users was as high as 68%, with a precision rate of up to 90%. This discovery challenges the assumption that pseudonymity provides adequate protection online. By linking posts across different platforms and using large language models, researchers were able to de-anonymize users and gather personal information, raising concerns about doxxing and stalking. The study highlights the need to reassess online privacy measures in the face of advancing AI technologies.