AI Breaks Internet Anonymity with New Deanonymization Technique
- Researchers from ETH Zurich and Anthropic developed an AI pipeline achieving a precision rate of up to 90% in matching anonymous accounts to real identities.
- The process relies on web search, embeddings, and reasoning models like GPT-5.2, costing approximately $1 to $4 per target without requiring hacking or data breaches.
- In tests, the AI correctly identified identities from pseudonymous Hacker News accounts linked to LinkedIn profiles in about two-thirds of cases.
- The study highlights the potential for AI-driven deanonymization using publicly available information without any data leaks.
This research underscores the vulnerability of online anonymity as AI tools can effectively identify individuals based on their public posts and profiles. The study’s findings emphasize the need for revised privacy threat models in light of these capabilities.
The cost-effective nature of this technique, which operates without breaching data security measures, presents significant implications for privacy online and raises concerns about how easily identities can be uncovered using advanced AI technologies. (Source)
