Discussion
Loading...

Post

Log in
  • Sign up
  • About
  • Code of conduct
  • Privacy
  • Users
  • Instances
  • About Bonfire
katzenberger
katzenberger
@katzenberger@tldr.nettime.org  ·  activity timestamp 6 days ago

Du hast ja sicher nichts zu verbergen.

»Wir zeigen, dass große Sprachmodelle für #Deanonymisierung im großen Maßstab eingesetzt werden können. Freien Internetzugang vorausgesetzt kann unser Agent Nutzer von #HackerNews und Teilnehmer von #AnthropicInterviews allein anhand von pseudonymen Online-Profilen und Onlinebeiträgen mit hoher Genauigkeit identifizieren, was bei einer manuellen Recherche Stunden kosten würde.

Anschließend entwickeln wir Angriffsmöglichkeiten für Plattformen, die nicht frei zugänglich sind. Wir implementieren eine eine skalierbare Angriffspipeline, basierend auf zwei Datenbanken mit unstrukturierten Texten, die über pseudonyme Individuen oder von ihnen geschrieben wurden.«

https://arxiv.org/abs/2602.16800

" #KI" #Datenschutz #Pseudonymität #Anonymität #LLM #LinkedIn

https://tldr.nettime.org/tags/Anonymit%C3%A4t
https://tldr.nettime.org/tags/Pseudonymit%C3%A4t
arXiv.org

Large-scale online deanonymization with LLMs

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. We then design attacks for the closed-world setting. Given two databases of pseudonymous individuals, each containing unstructured text written by or about that individual, we implement a scalable attack pipeline that uses LLMs to: (1) extract identity-relevant features, (2) search for candidate matches via semantic embeddings, and (3) reason over top candidates to verify matches and reduce false positives. Compared to classical deanonymization work (e.g., on the Netflix prize) that required structured data, our approach works directly on raw user content across arbitrary platforms. We construct three datasets with known ground-truth data to evaluate our attacks. The first links Hacker News to LinkedIn profiles, using cross-platform references that appear in the profiles. Our second dataset matches users across Reddit movie discussion communities; and the third splits a single user's Reddit history in time to create two pseudonymous profiles to be matched. In each setting, LLM-based methods substantially outperform classical baselines, achieving up to 68% recall at 90% precision compared to near 0% for the best non-LLM method. Our results show that the practical obscurity protecting pseudonymous users online no longer holds and that threat models for online privacy need to be reconsidered.
  • Copy link
  • Flag this post
  • Block

Indieweb Studio

This is a relaxed, online social space for the indieweb community, brought to you by indieweb.social.

Please abide by our code of conduct and have a nice time!

Indieweb Studio: About · Code of conduct · Privacy · Users · Instances
Bonfire social · 1.0.2-alpha.34 no JS en
Automatic federation enabled
Log in Create account
Instance logo
  • Explore
  • About
  • Members
  • Code of Conduct