EthereumETH, coin
DevelopingHack

ETH Zurich and MATS researchers report 90% precision de-anonymizing accounts

Decrypt reports on an ETH Zurich and MATS study that matched anonymous Hacker News accounts to real LinkedIn profiles at 90% precision using web search, embeddings and reasoning models.

Why it matters

If the method holds, anyone publishing under a pseudonym has a de-anonymization route that needs no database breach, which puts pressure on forums, regulators and platforms that host anonymous speech. Decrypt's single report, based on a February paper whose code and prompts were withheld, leaves the findings unreproduced so far.

Key facts

Researchers from ETH Zurich, MATS and Anthropic built an AI pipeline that matched anonymized Hacker News accounts to real LinkedIn profiles at 90% precision.

90 %Reported by Decrypt

The attack runs on web search, embeddings and reasoning models such as GPT-5.2, with no hacking or data breach involved.

Reported by Decrypt

The per-target cost of the attack is roughly $1 to $4.

Reported by Decrypt

The paper's authors withheld their code, prompts and every real identity they uncovered.

Reported by Decrypt

The study passed through ETH Zurich's ethics review board before publication.

Reported by Decrypt

The paper is titled 'Large-scale online deanonymization with LLMs' and comes from a February study.

Reported by Decrypt

Nicholas Carlini, a researcher at Anthropic, worked on the paper.

Reported by Decrypt

What happened

Decrypt reported on a research paper titled 'Large-scale online deanonymization with LLMs' by researchers at ETH Zurich and the AI safety group MATS, working with Nicholas Carlini of Anthropic. The researchers claim their pipeline matched anonymized Hacker News accounts to real LinkedIn profiles at 90% precision, using only web search, embeddings and reasoning models such as GPT-5.2, with no hacking or data breach involved. Decrypt put the per-target cost at roughly $1 to $4. The paper's authors withheld their code, prompts and every real identity they found, and say the study cleared ETH Zurich's ethics review board before publication. The report also notes the paper dates to a February study and that attention on X and Reddit came a day after its first report, with reaction from social media users described in the same article. No other outlet in this evidence set covers the work, so the claims rest on Decrypt's account alone.

How the story developed

  1. First report

    First report by Decrypt

    AI Can Now Doxx Your Anonymous Accounts? Here's What’s Going On

    Decrypt

  2. Update

    Study on LLM deanonymization completed

    The paper 'Large-scale online deanonymization with LLMs' dates from a February study by ETH Zurich, MATS and Anthropic researchers.

    Decrypt

  3. Update

    Decrypt reports the study

    Decrypt reported the paper and the 90% precision figure, noting the research was from February and that reaction circulated on X and Reddit.

    Decrypt

Affected entities

Original sources

PublisherReportRolePublished
Decrypt
Crypto media
AI Can Now Doxx Your Anonymous Accounts? Here's What’s Going OnOriginal