ASIRF: An Agentic Framework for Context-Dependent Sensitive Information Redaction

2026-09-24

AI Summary

This study addresses the challenge that definitions of sensitive information vary across domains, typically necessitating costly retraining in conventional systems. To this end, we propose ASIRF, a framework that introduces a novel agent-based dynamic knowledge retrieval mechanism. By integrating a multi-agent pipeline and a single-agent architecture with a flexible knowledge base, ASIRF dynamically adapts to domain-specific definitions during inference. It enables small open-source models to perform cross-domain data anonymization using minimal expert knowledge, thereby eliminating reliance on fixed taxonomies. Extensive experiments conducted across ten models and eight datasets demonstrate that ASIRF surpasses the OpenAI privacy filter baseline in recall for 85% of the evaluated model-domain combinations.

Abstract

Sensitive information is defined by domain and intent, not a universal category, yet redaction systems such as privacy filters and named-entity recognizers fix a taxonomy at training time, requiring retraining for each new domain. We introduce ASIRF (Agentic Sensitive Information Redaction Framework), which retrieves domain-specific definitions based on the input's domain from a flexible knowledge base at inference time, needing no retraining to adapt. Two architectures, a three-call multi-agent pipeline and a single-agent variant, are evaluated across ten small open-weight models and eight datasets, including out-of-distribution fictional domains, against the OpenAI Privacy Filter (OPF) as a trained-classifier baseline. With only a few dozen expert-authored definitions per domain and no training data, ASIRF's recall exceeds OPF's in 68 of 80 model-domain combinations (85 percent), by at least one of the two architectures, with shortfalls confined mostly to OPF's training-distribution domains.

Problem

• sensitive information redaction
• domain adaptation
• context-dependent privacy
• information filtering

Innovation

• Agentic Framework
• Sensitive Information Redaction
• Context-Dependent
• Multi-Agent Pipeline
• Zero-Shot Adaptation