A report co-authored by researchers from ETH Zurich, AI safety organization MATS, and Anthropic indicates that large language models (LLMs) can identify the real identities of anonymous users from their public posts through reading, searching, and reasoning. The study found that AI successfully identified 67% of anonymous Hacker News users and 9 interviewed scientists in tests, at a cost of approximately $1-4 per identification, raising market concerns about online privacy.