1 of 1
Story summary
- The study by Anthropic and ETH Zurich shows large language models (LLMs) can deanonymize online accounts by linking pseudonymous profiles to real identities.
- LLMs analyzing public text in the arXiv-published study achieve up to 68% recall and 90% precision, outperforming traditional methods.
- The cost to identify an account could be $1-$4 per profile, raising privacy concerns.
- Researchers call for stronger privacy tools and safeguards as AI capabilities grow.
