Your Online Anonymity Is Worth $2: How AI Can Unmask You

3 min read
Article

ETH Zurich researchers show an LLM can unmask an anonymous account for $1-4, with 90% accuracy. The economics of online anonymity just collapsed.

The free AI newsletter
Your Online Anonymity Is Worth $2: How AI Can Unmask You

That burner Reddit account. Your alt Twitter. The Glassdoor review where you torched your boss. You figured no one would ever connect the dots back to you. Turns out it costs two bucks.

The invisible contract

Online anonymity was never a guaranteed right. It was an economic handshake. Finding you cost too much time and effort to bother. Think of it like a house with no locks, sitting at the end of a 200-mile forest road. You're not getting robbed because your door's secure. You're safe because the trip isn't worth it. A human investigator could've pieced together your Reddit history and LinkedIn profile. But it would've taken hours, maybe days, for a single match. That math just broke.

Two dollars to find you

On February 20, 2026, ETH Zurich researchers Simon Lermen and Daniel Paleka (affiliated with Anthropic and MATS Berkeley) published "Large-scale online deanonymization with LLMs." Matching Hacker News accounts to LinkedIn profiles: 68% recall at 90% precision. Reddit to Netflix with 10+ shared films: 48% recall at 90% precision. Total cost: under $2,000. Per profile: $1 to $4. Lermen told The Verge: "The economics are totally different now."

How the AI connects you

The system runs in four steps. Extraction: the LLM scans your posts for identity micropatterns. Search: semantic embeddings hunt for matches. Reasoning: the LLM weighs evidence like an investigator building a case. Calibration: confidence scoring. Paleka notes "every single thing the LLM found in principle could be found by a human investigator." It's the difference between picking one lock and picking a million locks overnight.

What this means for you

$1 to $4 per profile means anyone can run mass deanonymization. Nosy employer, vindictive ex, authoritarian regime. What's at stake is speech itself. The chilling effect is real. Important caveat: this is a preprint, not yet peer-reviewed. Satoshi Nakamoto's still anonymous, Signal still protects your messages. But the threat model shifted. Used to require an intelligence agency. Tomorrow it's a teenager with a credit card and API access.

Topics covered:

PrivacyNews

Frequently asked questions

How can AI deanonymize an online account?
An LLM analyzes posts from an anonymous account to extract identity micropatterns (writing style, posting times, biographical details), then cross-references them with public profiles using semantic embeddings. The system reasons about matches and assigns a confidence score.
How much does AI deanonymization cost?
According to the ETH Zurich study published in February 2026, the cost is $1 to $4 per profile. The complete experiment (thousands of profiles) cost under $2,000.
How reliable is LLM-based deanonymization?
On Hacker News / LinkedIn matching, the system achieved 68% recall with 90% precision. On Reddit / Netflix with 10+ films in common, 48% recall at 90% precision.
The free AI newsletter