Agent framework improves sensitive data redaction without retraining
ASIRF: An Agentic Framework for Context-Dependent Sensitive Information Redaction
Artificial IntelligenceInformation RetrievalMachine Learning
Summary
Sensitive information varies depending on where it appears and why it matters, but current systems must be retrained for each new situation. The authors propose ASIRF, a system that uses expert rules about what counts as sensitive in each setting without needing to learn anew. ASIRF works through flexible knowledge and can handle many types of data, including fictional or unfamiliar ones. It outperforms a common trained filter in most tests without using any training examples.
What this means in practice
- •For data privacy teams: Adapt sensitive data redaction systems to new domains quickly without retraining by using domain-specific expert rules.
- •For software developers: Integrate context-aware redaction that dynamically adjusts to different application domains using flexible knowledge bases.
Authors
Sudha Priyadarshini, Mohamed Chahine Ghanem
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.