AI Deanonymization Method Identifies Users With 67% Accuracy
In brief
- AI deanonymization method identifies anonymous users by extracting personal details, searching matches, and cross-checking with reasoning models.
- Technique achieved 67% accuracy on Hacker News users and identified at least 9 scientists from interview transcripts.
- Attack costs $1–$4 in AI fees, requires no data breach, using only standard chatbot capabilities.
How the AI Attack Works
The researchers broke the attack into four steps: Extract, Search, Reason, and Calibrate. Extract means feeding an AI someone's posts and having it summarize the person: where they might live, their job, their hobbies, even a quirk like British spelling or a dog's name mentioned in passing. Search converts that summary into an embedding (a string of numbers a computer can use to instantly find similar-sounding profiles among thousands of candidates). Reason is where a more powerful model looks at the best few matches and thinks through which one actually fits, cross-checking details. Calibrate has the AI rate its own confidence, so it only guesses when it's fairly sure and stays quiet otherwise.
The cost is minimal. Running one of these searches costs between $1 and $4 in AI subscription fees, according to the paper.
What the Tests Showed
The results are sobering. In one test, the team gathered 338 users of the mostly pseudonymous Hacker News message board who had linked their LinkedIn profiles. The AI correctly identified 226 of them—a 67% success rate. The AI was wrong on roughly one in 10 of the guesses it actually made.
The Implications
The paper argues that pseudonymous users can no longer rely on practical obscurity. The researchers claim that large language models, the AI systems behind Claude, ChatGPT, and Gemini, can read someone's pseudonymous posts and determine their real-world identity without hacking. This matters because it upends how we think about online privacy.
"Our results show that the practical obscurity protecting pseudonymous users online no longer holds and that threat models for online privacy need to be reconsidered," the paper reads.
The study resurfaced on social media this week, months after its February publication. For anyone who relies on pseudonymity for safety, free speech, or professional reasons, the implications are worth taking seriously.
Frequently asked questions
How does the AI identify anonymous users?
The AI extracts personal details from someone's posts (location, job, hobbies, writing quirks), converts that summary into an embedding to search for matches, then uses a more powerful model to reason through which match fits best by cross-checking details.
How accurate is this method?
In tests on Hacker News users, the AI correctly identified 226 out of 338 users (67% accuracy) and was wrong on roughly one in ten guesses it actually made. In a separate test on scientist interviews, it correctly identified at least 9 people from their work descriptions.
Does this require hacking or a data breach?
No. The method uses only web search and summarization abilities built into ordinary chatbots, requiring no data breach, hack, or insider access. Each search costs between $1 and $4 in AI subscription fees.
