AI Large Models Can Identify Two-Thirds of Anonymous Users' Identities, Online Privacy Faces New Challenges

For a long time, anonymous accounts have been an important protection for internet users to avoid harassment and doxxing, but the rapid development of AI technology is weakening this protection. Researchers from ETH Zurich and Anthropic found that large language models can identify the real identities of anonymous users through forum posts and profiles, successfully recognizing about two-thirds of users in experiments. This de-anonymization ability not only applies to forum data but also works in scenarios like LinkedIn profiles and anonymous interview records, and can even identify 7% of users from daily surveys. Researchers warn that the current privacy assumptions of the internet are no longer valid, and users, platforms, and policymakers need to take stronger protective measures.