Study Finds LLMs Can De-anonymize Users at Scale, Threatening Privacy of Satoshi Nakamoto and Others

Online anonymity often relies on “practical obscurity”: accounts may not disclose real names, but cross-platform posts reveal occupations, locations, interests and writing styles that previously required time-consuming manual comparison. LLMs can now automatically extract these clues and search for potential identities, threatening whistleblowers and pseudonymous users. The crypto community also fears that articles, emails and forum posts left by Satoshi Nakamoto could be used to identify the Bitcoin creator, although the study neither established Nakamoto’s identity nor compromised the blockchain itself.

Researchers from ETH Zurich, Anthropic and MATS published a preprint on February 18, 2026, testing data from Hacker News, LinkedIn and Reddit. The LLM-based method recovered up to 68% of matches at 90% precision, while traditional methods achieved nearly 0%. The entire experiment cost less than $2,000, with each profile costing about $1–$4, or roughly NT$32–NT$128, and taking only minutes to process.