Breaking

Study Shows LLMs Can Deanonymize Pseudonymous Users at Up to 67% Recall

June 17, 2026 at 10:31 EDT

Security
Research & Papers
AI Agents

Researchers from ETH Zurich, Anthropic and MATS have published a paper demonstrating that LLMs can link pseudonymous accounts on Hacker News and Reddit to real identities with high accuracy. The work shows that large-scale "deanonymization" from post text alone is moving into practical reach at a cost of just a few dollars per profile.

February 2026 · ETH Zurich & Anthropic

LLMs Can Now Unmask Anonymous Posters — for About a Dollar

Autonomous AI agents link pseudonymous forum and Reddit posts to real profiles on sites like LinkedIn, automating an investigation that once took heavy manual effort — now reproducible in minutes.

67–68%

Recall matching Hacker News users to LinkedIn, at 90% precision

~$1–4

Estimated cost to deanonymize a single target

≈0%

Recall of prior non-LLM methods on the same task

The leap in re-identification

HN → LinkedIn recall at 90% precision — LLM agent vs. earlier methods

≈0%

Prior methods

67–68%

LLM agent

The four-stage attack pipeline

Extract

Pull attributes & "breadcrumbs" from posts

→

Semantic embedding search across the web

→

Reason

LLM verifies the candidate match

→

Calibrate

Score confidence in the identification

The concern

Identification for roughly a dollar invites misuse by stalkers and hackers
"Practical obscurity" of public data erodes — avoid persistent usernames
Platforms urged to revisit their privacy policies

The caveat

Evaluation used already-public, identifiable profiles
The most privacy-conscious users are less likely to be captured
Results should not be overstated

Why it matters

Unlike classic attacks on structured data, this works on unstructured text — forum posts, cross-Reddit history, throwaway accounts — turning a once-theoretical threat into a cheap, automated reality. The authors argue the threat model itself must be reconsidered.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.