New research from ETH Zurich, MATS and Anthropic shows that advanced AI systems can connect pseudonymous online profiles with real-world identities more efficiently than traditional search methods.
In controlled tests involving accounts whose identities were already publicly verifiable, an AI agent analyzed public posts after obvious identifiers such as names, usernames and profile links had been removed. The system combined subtle signals including professional interests, locations, hobbies, writing patterns and technical knowledge.
Small digital clues can form a larger profile
The researchers found that individual details may seem harmless in isolation. Yet when an AI model examines many posts at once, these fragments can create a distinctive digital pattern. This ability allows large language models to interpret unstructured language and identify meaningful connections across online activity.
In one experiment, the model correctly identified 226 out of 338 profiles while maintaining high precision. Additional tests involving academic and software-engineering communities produced varied results, showing that success depends on how much contextual information a person has shared publicly.
Privacy awareness enters a new AI era
The study also explored whether AI could link older and newer activity from the same Reddit users. Even after a year-long gap was removed from users' posting histories, the strongest system connected a notable share of accounts with 99% precision when it made a match.
Researchers describe this shift as a reduction in "practical obscurity": information that once required hours of skilled manual investigation can increasingly be reviewed by AI in minutes. The findings do not mean online anonymity has disappeared, but they underline the value of limiting unnecessary personal details across public profiles.
As AI becomes more capable of interpreting digital footprints, privacy-by-design habits and clearer platform safeguards may become essential elements of a more resilient online future.
