New benchmarks and methods for PII detection in LLMs

Arxiv CS.CL (NLP/language models)

Four arXiv papers propose new benchmarks (REDACT, RedactionBench) and methods (local AI cascade, disentangled representations) for PII detection and text sanitization. The papers highlight limitations in existing PII corpora and address challenges like domain-specific identifiers and distributed agent privacy.

Privacyredaction

See 3 more sources

Redact or Keep? A Fully Local AI Cascade for Educational Dialogue De-Identification5 days agoHaocheng Zhang, Zhuqian Zhou, Kirk Vanacore, Bakhtawar Ahtisham, Ren\'e F. Kizilcec

RedactionBench5 days agoSean Brynj\'olfsson, Shashvat Jayakrishnan, Esha Sali, Diptanshu Purwar, Madhav Aggarwal

Privacy-Preserving Text Sanitization for Distributed Agents Collaboration via Disentangled Representations7 days agoXuan Liu, Hefeng Zhou, Sicheng Chen, Chao Yang, Xingcheng Xu, Jingjing Qu, Jiong Lou, Jie LI, Xia Hu

Arxiv CS.CL (NLP/language models)·Guneesh VatsAnubha AgrawalShikha SinghalAjita DashPraison SelvarajVidhan JhawarRanga Prasad ChennaBharadwaj Y M G

Jun 19, 4:00 AM