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

Sensitive information is defined by domain and intent, not a universal category. Existing detectors answer this with training, either by learning a fixed category set or by learning about the notion of entities from named-entity supervision. We introduce ASIRF, an agentic framework that adapts to a new domain with no training run and no labeled data. It classifies the input’s domain, retrieves matching definitions from a knowledge base at inference time, and extracts the values. 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%), by at least one of the two architectures, with shortfalls confined mostly to OPF’s training-distribution domains.