Artificial Intelligence Cracks the Code of Pseudonymous Users
New research reveals that language models can unmask anonymity online with surprising accuracy, jeopardizing privacy and security for many social media users.
Artificial Intelligence is becoming increasingly adept at unmasking the identities behind pseudonymous social media accounts, posing a significant threat to user privacy, according to recent research. The findings suggest that language models can correlate specific individuals with their online personas across multiple platforms with a high degree of accuracy, potentially enabling a host of malicious activities such as doxxing and stalking.
The research, published in a recent paper, demonstrates that recall (the proportion of successfully deanonymized users) can reach as high as 68%, while precision (the accuracy of those identifications) can be up to 90%. This far exceeds the capabilities of traditional deanonymization techniques that rely on manual data gathering and algorithmic matching.
The implications of this study are far-reaching, as the ability to cheaply and quickly identify the people behind obscured accounts undermines a crucial privacy measure used by many to participate in sensitive public discussions. The assembly of detailed marketing profiles that track individuals' personal information, such as where they live and what they do for a living, is now a looming threat.
The researchers emphasize that this technology has the potential to
Source: Ars Technica
