Threats by #AI:
“We show that large language models can be used to perform
at-scale #deanonymization. With full Internet access, our agent
can re-identify Hacker News users and Anthropic Interviewer
participants at high precision, given #pseudonymous online
profiles and conversations alone, matching what would take
hours for a dedicated human investigator.”
“Deanonymization is a two-step process at heart, involving
profiling an anonymous person from their posts, and then
matching them to a known identity. It’s well-known that large
language models can infer personal attributes from
text on online forums. Given this, it makes sense to
ask: how good are LLMs at full end-to-end deanonymization,
and is this a practical threat to pseudonymous accounts?”
https://arxiv.org/pdf/2602.16800v2

