AI learns to uncover the identities of anonymous users
Researchers from ETH Zurich and Anthropic found that large language models can deanonymise social media users by analysing texts and indirect clues within their messages. This approach can identify users across different platforms, posing a significant risk to online privacy.
Source Incrypted
In experiments, the AI matched accounts across platforms with up to 90% accuracy in identification, successfully revealing 68% of users. Scientists warn that such technologies could facilitate mass deanonymisation and increase privacy risks, including doxing and intrusive hyper-targeted advertising.
This discovery highlights the urgent need to develop more robust personal data protection tools in the digital space, as threats to anonymity and privacy are likely to intensify.
