A recent study from ETH Zurich, Anthropic, and the Machine Learning Alignment program reveals how AI can effectively deanonymize online accounts. Researchers developed automated AI agents that outperform traditional methods in reidentifying anonymized material by analyzing text patterns, biographical clues, and posting behavior. The AI achieved correct identification of matching accounts up to 68% of the time with 90% precision, significantly outpacing existing techniques. While the study highlights concerns for online privacy, particularly for journalists and activists, it also emphasizes that anonymity isn’t completely dead. Ethical testing on real users was avoided, and researchers stress that basic safety measures should still be employed. Despite significant AI advancements, caution is warranted as these systems are not perfect in real-world applications. Users are advised to consider their posts carefully to maintain pseudonymity, while social platforms need to implement safeguards against data scraping to protect online identities.
Source link
Share
Read more