匿名时代的终结：大模型只需几篇帖子就能识破你的马甲
2026-03-04
煎蛋
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Recent research by Anthropic and ETH Zurich demonstrates that large language models (LLMs) can deanonymize online users with up to 90% accuracy by analyzing unstructured text across platforms. This AI-driven capability undermines online anonymity, enabling large-scale privacy violations and exposing users to tracking and profiling at minimal cost.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves the use of AI systems (large language models) to analyze user-generated text and identify real identities behind anonymous online accounts. This use of AI directly leads to violations of privacy and potentially breaches fundamental rights, which qualifies as harm under the framework. The article reports actual research results showing high accuracy in de-anonymization, implying that harm is occurring or imminent, not just a theoretical risk. Therefore, this constitutes an AI Incident due to realized harm to privacy and rights caused by AI use.[AI generated]