AI Can Unmask Anonymous Posters for a Few Dollars a Profile
Ourtake
Persistent pseudonyms may no longer offer meaningful protection against a determined adversary with cheap AI tools.
What happened
Researchers (Lermen et al.) found that large language models can match pseudonymous online profiles to real identities automatically, for about one to four dollars per profile. In one experiment an AI agent correctly matched about two-thirds of 338 Hacker News profiles to LinkedIn identities, with a false positive rate near ten percent. The researchers are not releasing attack code or identities, and say countermeasures are hard to find.