Pith. sign in

REVIEW 2 cited by

Automated Privacy Information Annotation in Large Language Model Interactions

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2505.20910 v2 pith:SZOPHXOL submitted 2025-05-27 cs.CL

Automated Privacy Information Annotation in Large Language Model Interactions

classification cs.CL
keywords privacyinformationevaluationllmsmethodsdetectionphrasesannotation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Users interacting with large language models (LLMs) under their real identifiers often unknowingly risk disclosing private information. Automatically notifying users whether their queries leak privacy and which phrases leak what private information has therefore become a practical need. Existing privacy detection methods, however, were designed for different objectives and application domains, typically tagging personally identifiable information (PII) in anonymous content, which is insufficient in real-name interaction scenarios with LLMs. In this work, to support the development and evaluation of privacy detection models for LLM interactions that are deployable on local user devices, we construct a large-scale multilingual dataset with 249K user queries and 154K annotated privacy phrases. In particular, we build an automated privacy annotation pipeline with strong LLMs to automatically extract privacy phrases from dialogue datasets and annotate leaked information. We also design evaluation metrics at the levels of privacy leakage, extracted privacy phrase, and privacy information. We further establish baseline methods using light-weight LLMs with both tuning-free and tuning-based methods, and report a comprehensive evaluation of their performance. Evaluation results reveal a gap between current performance and the requirements of real-world LLM applications, motivating future research into more effective local privacy detection methods grounded in our dataset.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. PIIGuard: Mitigating PII Harvesting under Adversarial Sanitization

    cs.CR 2026-05 conditional novelty 7.0

    PIIGuard uses optimized hidden HTML fragments on webpages to block LLMs from leaking contact PII via indirect prompt injection, achieving at least 97% defense success across tested models while preserving benign QA utility.

  2. Addressing Labelled Data Scarcity: Taxonomy-Agnostic Annotation of PII Values in HTTP Traffic using LLMs

    cs.AI 2026-05 unverdicted novelty 6.0

    LLMs support taxonomy-agnostic detection and value extraction of PII in HTTP traffic via a deterministic pre-processing plus classification pipeline, plus an LLM generator for synthetic labeled traffic.