Reproduce “Automated Profile Inference with Language Model Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuntao Du, Zitao Li, Bolin Ding, Yaliang Li, Hanshen Xiao, Jingren Zhou, Ninghui Li
Researchers introduced and evaluated AutoProfiler, a multi-agent framework leveraging large language models to autonomously infer detailed personal profiles, including Personally Identifiable Information (PII) and Sensitive Personal Information (SPI), from public online activities on pseudonymous platforms. The framework achieved 100% accuracy for high-confidence inferences and profiled users 120 times faster and 50 times cheaper than human profilers, demonstrating a new scalable privacy threat.
Key insights
- LLM-based agents can autonomously infer comprehensive personal profiles, including both PII and SPI, from public pseudonymous online activities with high accuracy and confidence.
- The inferred profiles, even from subtle and seemingly innocuous cues, pose a severe privacy threat, enabling de-anonymization and targeted malicious exploitation.
- Current anonymization tools and LLM alignment strategies are largely insufficient against this automated profile inference, as LLMs can leverage contextual rather than just direct identifiers.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.