Anonymity on the Net is No More: AI Finds Real People in Minutes

Illustrative photo. Photo: freepik.com

This conclusion was reached by researchers from the Swiss Federal Institute of Technology Zurich and Anthropic. The scientists showed that modern AI models are capable of automatically matching anonymous profiles with real people, using only message texts without structured data such as tables or ratings.

While similar investigations previously required hours of manual work, AI now performs this in a fully automated mode.

In one experiment, researchers took 338 users of the Hacker News portal who had previously linked to their LinkedIn profiles. Names and direct links were removed from the descriptions. After this, the AI agent was given only impersonal texts of publications and instructed to find the real identity of the account owner. The model correctly matched 226 out of 338 profiles with an accuracy of about 90%. In other words, the vast majority of attempts to guess a profile were correct.

The system works in stages: the language model analyzes messages and extracts features from them — profession, place of residence, interests, and other details. Based on this data, a profile is formed, which is automatically matched with possible candidates in open sources. Then the model checks the most probable matches and evaluates the level of confidence in the result.

Separately, researchers tested the method on public interviews of scientists about the application of AI in their work. Despite partial text editing, the agent was able to identify the identities of at least 9 out of 33 participants with high accuracy.

The researchers note that the problem is not that AI is "smarter" than a human investigator. The model uses the same indirect clues: mentions of a university, project, city, hobby. The difference lies in speed and cost, because what previously required significant effort and was therefore rarely used can now be scaled.

The authors warn that this changes fundamental perceptions of online anonymity. According to them, the "practical obscurity" that users relied on — the belief that they could theoretically be unmasked, but no one would bother — no longer works. The potential consequences of applying such a tool, researchers point out, are extremely broad: from the persecution of activists and journalists to mass targeted fraud.

• 1
• 2
• 1
• 3
• 6
• 1
• 5
• 0