Sullivan & Cromwell brings legal expertise into agreement review

OpenAI’s case study says Sullivan & Cromwell partner John Savva built Agreement Analyzer first as a Codex single-user proof of concept, then collaborated with OpenAI to improve quality, add parallel processing, and support multiple lawyers. The tool integrates firm workflows, curated precedents, negotiating playbooks, and ChatGPT Enterprise. Lawyers input a draft, add precedents and playbooks, and get a memorandum plus proposed tracked changes after multiple review cycles from different perspectives.

Savva emphasizes source transparency for market-practice claims and describes adversarial agent setups where specialized agents review in parallel and challenge each other’s findings. He points to GPT-6 Astra’s reasoning quality as unlocking deeper analysis. The framing: augment judgment, don’t replace lawyers.

Legal AI that cites its precedents and leaves tracked changes for humans is the adult form of the category. Adversarial agent review is a smart pattern for a profession that already thinks in argument and counterargument.

Hallucinated market practice remains the failure mode; Savva’s insistence on visible sources is the control. Useful if associates spend saved hours on judgment instead of first-pass redlines only.

I’m watching whether Agreement Analyzer stays an internal S&C edge or becomes a template other firms copy with their own playbooks.

Savva says AI use inside the firm has moved from administrative tasks toward deep analysis of complex agreements as model reasoning improved.