Benchmark RadarAI BENCHMARK PROFILE
ContractScrub
Evaluates LLMs on legal contract scrubbing tasks using hand-crafted contracts covering error categories like defined term misuse and inconsistent language.
- Released
- 2026-08-20
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Provides the first formal evaluation of contract scrubbing, revealing that frontier models perform surprisingly poorly on this specific legal review task despite strong general benchmarks.
Motivation
Legal work, with its heavy reliance on processing large amounts of text, is often considered one of the domains most exposed to the use of LLMs.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.