Benchmark RadarAI BENCHMARK PROFILE
RedactionBench
RedactionBench evaluates contextual redaction of PII across 200 documents and 11 domains, with a character-level R-Score metric that treats semantically similar redactions equally.
- Released
- 2026-06-17
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Existing redaction benchmarks conflate extraction with privacy semantics; RedactionBench introduces contextual integrity and a metric that decouples ambiguity from precision.
Motivation
Large Language Models are increasingly applied to sensitive domains that require redaction of personally identifiable information (PII).
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.