āļ„āļđāđˆāļĄāļ·āļ­āļ—āļēāļ‡āđ€āļ—āļ„āļ™āļīāļ„

CUAD āđāļĨāļ°āļŠāļļāļ”āļ‚āđ‰āļ­āļĄāļđāļĨ NLP āļ—āļēāļ‡āļāļŽāļŦāļĄāļēāļĒ

CUAD is an expert-annotated dataset for contract-review NLP, not a complete representation of all agreements or legal questions.

āļŸāļąāļ‡āļ„āļģāđāļ™āļ°āļ™āļģāļ™āļĩāđ‰āļšāļĢāļĢāļĒāļēāļĒāđ‚āļ”āļĒ AI · ~3 min

āļšāļ™āļŦāļ™āđ‰āļēāļ™āļĩāđ‰āļ­āđˆāļēāļ™ 3 āļ™āļēāļ—āļĩ
  1. āļ āļēāļžāļĢāļ§āļĄ
  2. āđ€āļˆāļēāļ°āļĨāļķāļ
  3. āļāļēāļĢāđƒāļŠāđ‰āļ‡āļēāļ™āļˆāļĢāļīāļ‡āđƒāļ™āđ‚āļĨāļāđāļŦāđˆāļ‡āļ„āļ§āļēāļĄāđ€āļ›āđ‡āļ™āļˆāļĢāļīāļ‡
  4. The Future of CUAD and Legal NLP Datasets
  5. āļāļēāļĢāļ™āļģāđ„āļ›āļ›āļāļīāļšāļąāļ•āļī
  6. āļŠāļģāļĢāļ§āļˆāļ•āđˆāļ­āđ„āļ›
  7. āļ„āļģāļ–āļēāļĄāļ—āļĩāđˆāļžāļšāļšāđˆāļ­āļĒ

āļ āļēāļžāļĢāļ§āļĄ

Its labels target 41 clause types in 510 commercial contracts; benchmark results measure performance on that task and distribution, not a system’s ability to practice law.

āđ€āļˆāļēāļ°āļĨāļķāļ

The Contract Understanding Atticus Dataset (CUAD) was created for natural-language processing research in contract review. The Atticus Project describes version 1 as 510 commercial contracts with more than 13,000 expert-supervised labels covering 41 clause types considered important in corporate transactions, including mergers and acquisitions. The dataset was accepted at NeurIPS 2021 and is released under a stated license on the project page. CUAD frames clause review as finding or classifying spans of contract text. It can support research on clause detection, extraction, and related methods, but labels represent selected categories and annotation choices. A strong score does not establish that a system understands every legal interaction, that it performs well on other contract populations, or that its output is ready for a transaction. Contract styles, jurisdictions, languages, clause definitions, scanning quality, and amendments can all differ from the dataset. Before using CUAD, check its version, license, task definition, split design, and label handbook. Avoid leakage between training and test examples that share templates or related documents. Evaluate on separate agreements reflecting the intended use, inspect false positives and missed clauses, and have legal reviewers assess whether extracted spans preserve exceptions and context. CUAD is a research benchmark, not legal advice, a model certification, or an endorsement of automated contract review. Annotation instructions and adjudication also shape what counts as a correct span, so an evaluation should match the dataset’s exact task definition.

āļāļēāļĢāđƒāļŠāđ‰āļ‡āļēāļ™āļˆāļĢāļīāļ‡āđƒāļ™āđ‚āļĨāļāđāļŦāđˆāļ‡āļ„āļ§āļēāļĄāđ€āļ›āđ‡āļ™āļˆāļĢāļīāļ‡

A researcher evaluates a clause-finding model against CUAD’s annotated spans.

A reviewer checks whether a benchmark result transfers from commercial contracts to a new contract type.

A team studies which clause labels are represented before designing a contract-review model.

A data scientist separates development examples from held-out evaluation contracts.

The Future of CUAD and Legal NLP Datasets

Future contract benchmarks may add document types, languages, and richer measures of clause interaction. Such expansion can improve coverage but will not make a benchmark equivalent to legal practice. Teams should test against current agreements and keep experts involved in defining and validating outputs. Newer datasets may broaden contract types or model clause relationships, yet deployment still needs target-specific evaluation. Compare results against the actual agreements and business decisions in scope, and keep the model from silently converting benchmark labels into legal conclusions.

āļāļēāļĢāļ™āļģāđ„āļ›āļ›āļāļīāļšāļąāļ•āļī

āļ—āļģāđ„āļĄāļĄāļąāļ™āļ–āļķāļ‡āļŠāļģāļ„āļąāļ

  • āļ•āđ‰āļ™āļ—āļļāļ™āđāļĨāļ°āļ‡āļšāļ›āļĢāļ°āļĄāļēāļ“. āļāļēāļĢāļ•āļąāļ”āļŠāļīāļ™āđƒāļˆāļ”āđ‰āļēāļ™āļŠāļ–āļēāļ›āļąāļ•āļĒāļāļĢāļĢāļĄāļ‚āļąāļšāđ€āļ„āļĨāļ·āđˆāļ­āļ™āļ›āļĢāļ°āļŠāļīāļ—āļ˜āļīāļ āļēāļžāđāļĨāļ°āļ•āđ‰āļ™āļ—āļļāļ™āļāļēāļĢāļ”āļģāđ€āļ™āļīāļ™āļ‡āļēāļ™āđ€āļ›āđ‡āļ™āđ€āļ§āļĨāļēāļŦāļĨāļēāļĒāļ›āļĩ
  • āļāļēāļĢāļ•āļąāļ”āļŠāļīāļ™āđƒāļˆāļ—āļĩāđˆāļŠāļąāļ”āđ€āļˆāļ™āļĒāļīāđˆāļ‡āļ‚āļķāđ‰āļ™. āļāļēāļĢāļĻāļķāļāļĐāļēāļ”āđ‰āļēāļ™āđ€āļ—āļ„āļ™āļīāļ„āļŠāđˆāļ§āļĒāđƒāļŦāđ‰āļ—āļĩāļĄāđ€āļĨāļ·āļ­āļāļāļĨāļļāđˆāļĄāļ—āļĩāđˆāđ€āļŦāļĄāļēāļ°āļŠāļĄ āđ„āļĄāđˆāđƒāļŠāđˆāđāļ„āđˆāļāļĨāļļāđˆāļĄāđƒāļŦāļĄāđˆāļĨāđˆāļēāļŠāļļāļ”
  • āļāļēāļĢāļ„āļ§āļšāļ„āļļāļĄāļ„āļļāļ“āļ āļēāļž. āļ•āļąāļ§āđ€āļĨāļ·āļ­āļāļ—āļēāļ‡āļ§āļīāļĻāļ§āļāļĢāļĢāļĄāļ—āļĩāđˆāļ”āļĩāļāļ§āđˆāļēāļˆāļ°āļŠāđˆāļ§āļĒāļĨāļ”āđ€āļŦāļ•āļļāļāļēāļĢāļ“āđŒāļ”āđ‰āļēāļ™āļ„āļ§āļēāļĄāļ™āđˆāļēāđ€āļŠāļ·āđˆāļ­āļ–āļ·āļ­āđƒāļ™āļāļēāļĢāļœāļĨāļīāļ•

āļĢāļ°āļ§āļąāļ‡

  • āļāļēāļĢāđ€āļžāļīāđˆāļĄāļ›āļĢāļ°āļŠāļīāļ—āļ˜āļīāļ āļēāļžāđ€āļāļ“āļ‘āđŒāļĄāļēāļ•āļĢāļāļēāļ™āļŦāļ™āļķāđˆāļ‡āļĢāļēāļĒāļāļēāļĢāļŠāļēāļĄāļēāļĢāļ–āļ‹āđˆāļ­āļ™āļˆāļļāļ”āļ­āđˆāļ­āļ™āļ‚āļ­āļ‡āļĢāļ°āļšāļšāđƒāļ™āļ§āļ‡āļāļ§āđ‰āļēāļ‡āđ„āļ”āđ‰
  • āļ•āđ‰āļ™āļ—āļļāļ™āđ‚āļ„āļĢāļ‡āļŠāļĢāđ‰āļēāļ‡āļžāļ·āđ‰āļ™āļāļēāļ™āđāļĨāļ°āļāļēāļĢāļšāļģāļĢāļļāļ‡āļĢāļąāļāļĐāļēāļĄāļąāļāļ–āļđāļāļ›āļĢāļ°āđ€āļĄāļīāļ™āļ•āđˆāļģāđ„āļ›
  • āļŠāđˆāļ­āļ‡āļ§āđˆāļēāļ‡āļ”āđ‰āļēāļ™āļ„āļ§āļēāļĄāļ›āļĨāļ­āļ”āļ āļąāļĒāđāļĨāļ°āļ„āļ§āļēāļĄāļŠāļēāļĄāļēāļĢāļ–āđƒāļ™āļāļēāļĢāļŠāļąāļ‡āđ€āļāļ•āļŠāļēāļĄāļēāļĢāļ–āđ€āļžāļīāđˆāļĄāļ‚āļķāđ‰āļ™āđ„āļ”āđ‰āđ€āļĄāļ·āđˆāļ­āļĢāļ°āļšāļšāļĄāļĩāļ„āļ§āļēāļĄāļ‹āļąāļšāļ‹āđ‰āļ­āļ™āļĄāļēāļāļ‚āļķāđ‰āļ™

āļ‚āļąāđ‰āļ™āļ•āļ­āļ™āļ•āđˆāļ­āđ„āļ›

  1. āļāļģāļŦāļ™āļ”āđ€āļ›āđ‰āļēāļŦāļĄāļēāļĒāđ€āļ§āļĨāļēāđāļāļ‡ āļ„āļļāļ“āļ āļēāļž āđāļĨāļ°āļ•āđ‰āļ™āļ—āļļāļ™āļāđˆāļ­āļ™āļ™āļģāđ„āļ›āđƒāļŠāđ‰āļ‡āļēāļ™
  2. āđ€āļāļ“āļ‘āđŒāļĄāļēāļ•āļĢāļāļēāļ™āļ āļēāļĒāđƒāļ•āđ‰āļŠāļ āļēāļ§āļ°āđ‚āļŦāļĨāļ”āđāļĨāļ°āļ‚āđ‰āļ­āļĄāļđāļĨāļˆāļĢāļīāļ‡
  3. āļāļēāļĢāļ•āļĢāļ§āļˆāļŠāļ­āļšāđ€āļ„āļĢāļ·āđˆāļ­āļ‡āļĄāļ·āļ­āđ€āļžāļ·āđˆāļ­āļŦāļēāļ‚āđ‰āļ­āļœāļīāļ”āļžāļĨāļēāļ” āļāļēāļĢāđ€āļšāļĩāđˆāļĒāļ‡āđ€āļšāļ™ āđāļĨāļ°āļœāļĨāļāļĢāļ°āļ—āļšāļ•āđˆāļ­āļœāļđāđ‰āđƒāļŠāđ‰
  4. āđ€āļ•āļĢāļĩāļĒāļĄāđ€āļŠāđ‰āļ™āļ—āļēāļ‡āļāļēāļĢāļĒāđ‰āļ­āļ™āļāļĨāļąāļšāđāļĨāļ°āļāļēāļĢāļ•āļ­āļšāļŠāļ™āļ­āļ‡āļ•āđˆāļ­āđ€āļŦāļ•āļļāļāļēāļĢāļ“āđŒāļāđˆāļ­āļ™āļ›āļĢāļąāļšāļ‚āļ™āļēāļ”

āļŠāļģāļĢāļ§āļˆāļ•āđˆāļ­āđ„āļ›

āļˆāļ”āļŦāļĄāļēāļĒāļ‚āđˆāļēāļ§āļŸāļĢāļĩ

āļĢāļąāļšāļŠāļĢāļļāļ› AI āļĢāļēāļĒāļ§āļąāļ™

āđ€āļĢāļ·āđˆāļ­āļ‡āļĢāļēāļ§ AI āļ—āļĩāđˆāļ•āļĢāļ§āļˆāļŠāļ­āļšāđāļĨāđ‰āļ§ 3 āđ€āļĢāļ·āđˆāļ­āļ‡āļ—āļļāļāļ§āļąāļ™āļ—āļģāļ‡āļēāļ™ āđ€āļŠāđ‰āļē, āđ€āļ‚āļĩāļĒāļ™āđ€āļ›āđ‡āļ™āļ āļēāļĐāļēāļ­āļąāļ‡āļāļĪāļĐāļ˜āļĢāļĢāļĄāļ”āļē āļŸāļĢāļĩāļ•āļĨāļ­āļ”āđ„āļ› āđ„āļĄāđˆāļĄāļĩāđ‚āļ†āļĐāļ“āļē

āļ­āļĩāđ€āļĄāļĨāļŦāļ™āļķāđˆāļ‡āļ‰āļšāļąāļšāļ—āļļāļāļ§āļąāļ™āļ—āļģāļ‡āļēāļ™ āļĒāļāđ€āļĨāļīāļāļāļēāļĢāļŠāļĄāļąāļ„āļĢāđƒāļ™āļ„āļĨāļīāļāđ€āļ”āļĩāļĒāļ§ āđ€āļĢāļēāđ„āļĄāđˆāđ€āļ„āļĒāļ‚āļēāļĒāļŦāļĢāļ·āļ­āđāļŠāļĢāđŒāļ—āļĩāđˆāļ­āļĒāļđāđˆāļ­āļĩāđ€āļĄāļĨāļ‚āļ­āļ‡āļ„āļļāļ“

āļ—āļ”āļŠāļ­āļšāļ•āļąāļ§āđ€āļ­āļ‡

Take the CUAD and Legal NLP Datasets quiz

āļ‚āđ‰āļ­āđ€āļŠāļ™āļ­āđāļ™āļ°āļ—āļąāļ™āļ—āļĩāļŠāļģāļŦāļĢāļąāļšāļ—āļļāļāļ„āļģāļ•āļ­āļš āđāļĨāļ°āđƒāļšāļĢāļąāļšāļĢāļ­āļ‡āļ—āļĩāđˆāļŠāļēāļĄāļēāļĢāļ–āđāļŠāļĢāđŒāđ„āļ”āđ‰āļžāļĢāđ‰āļ­āļĄ ID āļ—āļĩāđˆāļ•āļĢāļ§āļˆāļŠāļ­āļšāđ„āļ”āđ‰āđ€āļĄāļ·āđˆāļ­āļ„āļļāļ“āļœāđˆāļēāļ™āļŦāļĨāļąāļāļŠāļđāļ•āļĢ

āđ€āļĢāļīāđˆāļĄāđāļšāļšāļ—āļ”āļŠāļ­āļš

āļŠāļ™āļąāļšāļŠāļ™āļļāļ™āļāļēāļĢāļĻāļķāļāļĐāļē AI āļŸāļĢāļĩ AI Understanding āđ€āļ›āđ‡āļ™āļ­āļ‡āļ„āđŒāļāļĢāđ„āļĄāđˆāđāļŠāļ§āļ‡āļŦāļēāļœāļĨāļāļģāđ„āļĢ 501(c)(3) — āđ„āļĄāđˆāļĄāļĩāđ‚āļ†āļĐāļ“āļē āđ„āļĄāđˆāļĄāļĩāļāļģāđāļžāļ‡āđ€āļ‡āļīāļ™, āļ•āļĨāļ­āļ”āđ„āļ› āļšāļĢāļīāļˆāļēāļ„

āļ„āļģāļ–āļēāļĄāļ—āļĩāđˆāļžāļšāļšāđˆāļ­āļĒ

What is CUAD and Legal NLP Datasets?

CUAD is an expert-annotated dataset for contract-review NLP, not a complete representation of all agreements or legal questions. Its labels target 41 clause types in 510 commercial contracts; benchmark results measure performance on that task and distribution, not a system’s ability to practice law.

How does the Atticus Project describe CUAD v1?

The project describes its size, labels, clause categories, and expert supervision.

Which task was CUAD created to support?

The paper presents CUAD as an expert-annotated dataset for contract review NLP.

A model scores highly on CUAD. What does that establish most directly?

A benchmark score is bounded by the dataset, task, and evaluation protocol.

Why examine the split design before using CUAD for evaluation?

Overlapping templates or related documents can create leakage between training and evaluation.

A lawyer wants to use a CUAD-trained model on a non-disclosure agreement in another jurisdiction. What is the best next step?

Different contract types and jurisdictions can shift language and task requirements.

āđ€āļĢāļĩāļĒāļ™āļĢāļđāđ‰āļ•āđˆāļ­āđ„āļ›

āļ„āļģāđāļ™āļ°āļ™āļģāļ—āļĩāđˆāđ€āļāļĩāđˆāļĒāļ§āļ‚āđ‰āļ­āļ‡

āļ„āļģāđāļ™āļ°āļ™āļģāđ€āļžāļīāđˆāļĄāđ€āļ•āļīāļĄāļ—āļĩāđˆāđ€āļĨāļ·āļ­āļāļŠāļģāļŦāļĢāļąāļšāļŦāļąāļ§āļ‚āđ‰āļ­āļ™āļĩāđ‰

āļ•āđˆāļ­āđ„āļ›āļ„āļģāđāļ™āļ°āļ™āļģāļ•āđˆāļ­āđ„āļ›

LegalBench and Evaluating AI on Legal Tasks

āđ€āļ—āļ„āļ™āļīāļ„

āļĢāļąāļšāđƒāļšāļĢāļąāļšāļĢāļ­āļ‡ AI āļŸāļĢāļĩāļŦāļĨāļąāļāļŠāļđāļ•āļĢāđāļĨāļ°āļāļēāļĢāļŠāļ­āļšāļŠāļąāđ‰āļ™ āđ† āļŸāļĢāļĩ āđ„āļĄāđˆāļĄāļĩāļ„āđˆāļēāđƒāļŠāđ‰āļˆāđˆāļēāļĒ āđ„āļĄāđˆāļĄāļĩāļšāļąāļ•āļĢāđ€āļ„āļĢāļ”āļīāļ• āļ›āļĢāļ°āļĄāļēāļ“ 2.5 āļŠāļąāđˆāļ§āđ‚āļĄāļ‡āđ€āļĢāļīāđˆāļĄāļ•āđ‰āļ™āļŸāļĢāļĩ →