—
LLMs can unmask pseudonymous users at scale with surprising accuracy
Researchers have demonstrated that large language models can deanonymize pseudonymous social media users across platforms with recall rates as high as 68% and precision up to 90%. The technique significantly outperforms classical deanonymization methods that relied on structured datasets or manual investigative work. The findings have broad privacy implications, as many users rely on pseudonymous accounts for sensitive discussions, and this capability opens individuals to doxxing, stalking, and