AI-Powered Deanonymization Threatens Online Anonymity

A new study by researchers at Anthropic and ETH Zurich has highlighted the growing threat of AI-powered deanonymization tools to online anonymity.

The system, which uses a multi-step process to analyze writing patterns, behaviors, and personal signals, was tested on 338 Hacker News users and achieved an impressive 90% accuracy rate in re-identifying them.

When applied to Reddit users across a one-year gap, the AI delivered a 450x improvement over traditional techniques, with an identification rate of 67.3% at 90% precision.

This capability poses significant risks to cryptocurrency and Web3 users who rely on pseudonyms to separate their financial activity from personal identity.

The researchers found that active participants in online discussions are more exposed than casual users, with a 48.1% identification rate among those discussing ten or more topics.