Sullivan & Cromwell and OpenAI have introduced Agreement Analyzer, a contract-review application that folds firm knowledge and client priorities into a single, iterative workflow for lawyers. Agreement Analyzer combines curated firm precedents, negotiating playbooks and ChatGPT Enterprise to surface potential issues in draft agreements, cite supporting authorities and generate proposed markups and a memorandum for attorneys to inspect and refine.
The tool is designed to begin with a lawyer uploading a draft agreement and selecting relevant firm precedents and a playbook that reflects both firm and client positions and prevailing market practice. Agreement Analyzer then conducts multiple review cycles from different analytical perspectives, synthesizes the outputs, and returns a consolidated set of identified issues, supporting explanations, cited precedents and proposed tracked changes. The approach is explicitly iterative and human-centered: users can accept, adapt or reject the tool’s edits as part of normal review practice.
A central design objective for Agreement Analyzer is transparency about sources and confidence. Outputs explicitly link findings to their underlying authorities so lawyers can verify the basis for claims about precedent or market norms. The application differentiates findings that survive internal challenge from those that are dismissed or left qualified, enabling lawyers to assess confidence levels before making final changes to a markup.
Sullivan & Cromwell partner John Savva described an adversarial review architecture that underpins this transparency. In that model, specialized agents generate conclusions while other agents test and challenge them. Parallel agents assess areas such as market deviations, risk allocation, clause interactions and potential ambiguities. Findings that remain supported after this internal challenge are labeled accordingly; disputed items are flagged; and unverified issues are retained with qualifications for human review. This structure is intended to make uncertainty and disagreement visible rather than hide it behind a single output.
From a development perspective, Agreement Analyzer began as a single-user prototype built by Savva with Codex. Sullivan & Cromwell then collaborated with OpenAI to improve output quality, increase processing speed through parallelization and evolve the prototype into a multiuser tool that can be used by several lawyers at once. The collaboration also focused on integrating the firm’s internal workflows with curated precedents, negotiating playbooks and ChatGPT Enterprise so that comparative analysis and authority-finding are largely automated rather than manually performed.
Agreement Analyzer’s workflow emphasizes both automation and lawyer oversight. By automating repetitive comparison and research tasks, the tool aims to free practitioners to concentrate on judgment-intensive questions where legal expertise and client context matter most. The system produces not only proposed tracked changes but also a supporting memorandum that documents the rationale and cites the precedents used in generating recommendations.
The application is built to be iterative: lawyers can use the outputs as a starting point, refining or rejecting suggestions as appropriate. That human-in-the-loop posture is central to Sullivan & Cromwell’s stated goal of improving efficiency without abdicating professional responsibility. Savva has emphasized that the technology is not intended to replace lawyers but to enable deeper analyses and higher-quality outputs for clients by directing human attention where judgment is required.
Looking ahead, Sullivan & Cromwell plans to continue exploring tools tailored to discrete legal tasks. Savva pointed to advances in reasoning with newer models, specifically citing GPT-6 Astra, as improving the quality of analytic output. The firm views such advances as key enablers for supporting more of the research and drafting workflow, while retaining lawyer review and final decision-making.
Agreement Analyzer illustrates a broader trend in legal technology: combining firm-specific knowledge assets with large-model reasoning to speed routine work and enhance the lawyer’s ability to focus on complex, judgment-driven issues. By making sources and uncertainty visible and by providing a mechanism for adversarial internal review, the tool seeks to balance automation with accountability. For now, Sullivan & Cromwell presents Agreement Analyzer as a practical instrument for enhancing contract review rather than a replacement for professional legal judgment.
Source: Read the original source
Leave a Reply