RedacBench: Can AI Erase Your Secrets?
Not on the map. Why?
Abstract (arXiv)
Modern language models can readily extract sensitive information from unstructured text, making redaction -- the selective removal of such information -- critical for data security. However, existing benchmarks for redaction typically focus on predefined categories of data such as personally identifiable information (PII) or evaluate specific techniques like masking. To address this limitation, we introduce RedacBench, a comprehensive benchmark for evaluating policy-conditioned redaction across domains and strategies. Constructed from 514 human-authored texts spanning individual, corporate, and government sources, paired with 187 security policies, RedacBench measures a model's ability to selectively remove policy-violating information while preserving the original semantics. We quantify performance using 8,053 annotated propositions that capture all inferable information in each text. This enables assessment of both security -- the removal of sensitive propositions -- and utility -- the preservation of non-sensitive propositions. Experiments across multiple redaction strategies and state-of-the-art language models show that while more advanced models can improve security, preserving utility remains a challenge. To facilitate future research, we release RedacBench along with a web-based playground for dataset customization and evaluation. Available at https://hyunjunian.github.io/redaction-playground/.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 21 of its 28 references that are in the map. First posted on arXiv 2 Mar 2026.
Builds on
21 of 28 references.
References · 21
- 2020Language Models are Few-Shot Learnerscited by 32,199
- 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
- 2020Extracting Training Data from Large Language Modelscited by 2,144
- 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
- 2021Differentially Private Fine-tuning of Language Modelscited by 392
- 2023Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theorycited by 218
- 2023Beyond Memorization: Violating Privacy Via Inference with Large Language Modelscited by 167
- 2023Emergent and Predictable Memorization in Large Language Modelscited by 147
- 2023Privacy-Preserving In-Context Learning with Differentially Private Few-Shot Generationcited by 94
- 2023DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineercited by 56
- 2023Privacy-Preserving In-Context Learning for Large Language Modelscited by 50
- 2024A Synthetic Dataset for Personal Attribute Inferencecited by 35
9 more
- 2024Teach LLMs to Phish: Stealing Private Information from Language Modelscited by 29
- 2025A False Sense of Privacy: Evaluating Textual Data Sanitization Beyond Surface-level Privacy Leakagecited by 15
- 2025Does Training with Synthetic Data Truly Protect Privacy?cited by 12
- 2023Text Sanitization Beyond Specific Domains: Zero-Shot Redaction & Substitution with Large Language Modelscited by 10
- 2025PRvL: Quantifying the Capabilities and Risks of Large Language Models for PII Redactioncited by 7
- 2024RedactBuster: Entity Type Recognition from Redacted Documentscited by 4
- 2025Adaptive PII Mitigation Framework for Large Language Modelscited by 4
- 2025Self-Refining Language Model Anonymizers via Adversarial Distillationcited by 4
- 2025Improving Privacy Benefits of Redactioncited by 2
Built on by
Citing papers per year: 2026: 1.
1 paper.