We measure the capabilities of LLMs to deanonymize users online.

They were able to de-anonymize posters from collections with 10,000’s of users, tested with scrapings from Hacker News, LinkedIn, and Reddit. Not good from a privacy perspective.