Sullivan & Cromwell partners with OpenAI on AI agreement-review tool

Law firm Sullivan & Cromwell has joined forces with OpenAI to create Agreement Analyzer, an artificial-intelligence tool that reviews draft contracts and flags potential issues.

John Savva, a partner at the firm, first built Agreement Analyzer as a single-user proof of concept using Codex. He then worked with OpenAI to improve the quality of the tool’s outputs, speed it up with parallel processing, and turn it into a system that multiple lawyers can use at once ¹.

The tool pulls together Sullivan & Cromwell’s own workflows, curated case precedents, negotiating playbooks and ChatGPT Enterprise ¹. A lawyer inputs a draft agreement and adds selected precedents, a playbook showing firm and client positions, and notes on market practice. The system then runs several review cycles from different angles, synthesizes the results and returns a memorandum along with proposed tracked changes for the lawyer to inspect, modify or reject .

How the tool works

Agreement Analyzer uses an adversarial setup in which different AI agents develop conclusions while other agents challenge them ¹. One group of specialized agents looks for market deviations and risk-allocation problems, another checks how clauses interact, and a third searches for ambiguities that could lead to disputes .

The output separates findings into three categories: results that survived challenge by other agents, results that were dismissed, and issues that remain unverified and are marked with qualifications ¹. Lawyers can read each finding’s explanation and see the supporting precedents before deciding whether to change the final document .

Savva said the aim is not to replace lawyers but to give them a tool that can dig deeper into complex agreements and surface more authority and precedent than a human reviewer could easily find alone ¹.

Why transparency matters

Trust in the tool depends on visibility. Savva stressed that if the AI makes a claim about market practice, a lawyer must be able to trace that claim back to its source ¹. The adversarial review process is designed to make that possible, so lawyers can see which conclusions are solid and which are still uncertain.

Looking ahead

Sullivan & Cromwell plans to keep building tools for specific legal tasks. Savva pointed to the newer reasoning abilities in GPT-6 Astra, saying those advances help remove a long-standing bottleneck in legal research and writing ¹.

Key facts

• Sullivan & Cromwell and OpenAI developed Agreement Analyzer to review contracts.
• The tool uses an adversarial AI setup where agents challenge each other's conclusions.
• Lawyers input precedents and playbooks so the tool reflects firm-specific guidance.
• John Savva credited GPT-6 Astra's reasoning improvements for advancing legal AI use.

Sources

1. Sullivan & Cromwell brings legal expertise into agreement review