OpenAI and Sullivan & Cromwell build Agreement Analyzer for contract review

OpenAI and Sullivan & Cromwell developed Agreement Analyzer, an internal AI application that reviews draft agreements, flags potential issues and proposes revisions for lawyers to inspect. The workflow packages parts of the law firm’s agreement-review process with selected precedents and negotiating playbooks, while lawyers decide whether to accept, adapt or reject its output.

The application combines ChatGPT Enterprise with Sullivan & Cromwell workflows and firm-selected material. According to the collaborators, it compares contract provisions with playbook positions over multiple review cycles before producing a memorandum or proposed revisions. Lawyers can inspect supporting explanations and relevant precedents before deciding whether to use a finding or change. That human-review structure parallels another Codex-built professional validation workflow covered by DataPhoenix, although the systems handle different documents.

Sullivan & Cromwell said partner John Savva first built a single-user proof of concept with Codex. OpenAI engineers then worked with the firm to refine the prototype and prepare it for use by more lawyers. The firm said Agreement Analyzer was developed for internal use; neither collaborator announced public access or pricing.

The firm said it unveiled and demonstrated the tool at OpenAI’s legal-industry summit in New York on September 17. OpenAI’s Astra for Law announcement from the same day also described Agreement Analyzer.

In Sullivan & Cromwell’s account of the project, Savva framed the near-term benefit as enabling a more thorough analysis within the available time, rather than necessarily shortening the review. The official materials do not provide an Agreement Analyzer-specific accuracy or error rate, a controlled performance comparison, a user count or an independently audited result. They also do not say whether broader deployment across the firm is complete.