AI Can Unmask Anonymous Users at Scale

Anthropic, ETH Zurich Study Shows Pseudonyms Offer Little Protection Rashmi Ramesh (rashmiramesh_) • March 2, 2026

Large language models can deanonymize pseudonymous online accounts automatically, cheaply and at scale, showed a study by researchers at ETH Zurich and Anthropic, challenging a foundational assumption of how online privacy works.

See Also: Why HSMs Are Critical to Digital Asset Security

Researchers correctly matched 67% of Hacker News users to their real LinkedIn profiles from a pool of 89,000 candidates, after removing all direct identifiers such as names, URLs and social handles. The entire experiment cost less than $2,000, with each online account costing between one and four dollars to identify.

Lead researcher Daniel Paleka told Information Security Media Group he found it surprising "how little information it takes to connect two accounts."

Co-researcher Simon Lermen said in a Substack post: "Ask yourself: could a team of smart investigators figure ...