SafeLawBench introduces a legal-safety taxonomy and 25,966 tasks on which 20 LLMs average 68.8% accuracy and the best model, Claude-3.5-Sonnet, reaches 80.5%.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SafeLawBench: Towards Safe Alignment of Large Language Models
SafeLawBench introduces a legal-safety taxonomy and 25,966 tasks on which 20 LLMs average 68.8% accuracy and the best model, Claude-3.5-Sonnet, reaches 80.5%.