CUAD

CUAD is a curated dataset of 13,000+ annotations designed to train AI models for specialized legal contract analysis.

Developed by The Atticus Project in collaboration with legal experts and data scientists, the Contract Understanding Atticus Dataset (CUAD) provides a high-quality corpus for training machine learning models to identify 41 distinct label types in commercial agreements. The dataset consists of 510 complex contracts manually reviewed by legal professionals to ensure precision in identifying clauses like change of control, indemnification, and termination rights. By open-sourcing this benchmark, the project enables developers to build NLP tools that automate the extraction of critical obligations and risks from dense legal documentation.

https://www.atticusproject.org/cuad