Law firm Ropes & Gray is developing a proprietary AI tool for transactional diligence in partnership with OpenAI, according to a case study published on OpenAI's website on October 29, 2026. The initial development has focused on material commercial contracts. According to the article, the tool combs through thousands of pages, identifies relevant documents and groups them by themes and issues, then produces summaries and a completed report in the firm's standard format. Howard Glazer, Partner and Head of Practice AI, is quoted saying clients "don't pay us to review contracts. They pay us to make judgments based on what's in the contracts." The firm compared test outputs against human-authored diligence reports from completed deals, then refined the rubric and report format. Another leader, identified as Lichtenstein, is quoted saying "Our lawyers still need to review everything," because the firm must be able to stand behind anything delivered to a client. The article describes the tool's output as comparable to a capable junior-to-mid-level associate and serving as a first pass. Ropes & Gray has publicly described an ambition to produce a detailed, issue-level report in hours rather than weeks. The firm has roughly 340 partners and 1,500 lawyers. The account comes from OpenAI, a party to the partnership; no independent performance figures are provided.
- Ropes & Gray is developing a proprietary diligence AI tool with OpenAI, per an OpenAI case study dated October 29, 2026
- The tool inventories and groups data-room documents, then drafts summaries and a report in the firm's format
- The firm states lawyers review all interim summaries and source documents before delivery to clients
- Stated ambition: produce an issue-level diligence report in hours rather than weeks (no verified metrics provided)
What it means for you
A large law firm has built a tool that drafts a first pass of document review, then has humans check it — the firm is explicit that lawyers review everything before anything reaches a client. The useful lesson here isn't the tool; it's the approach. They mapped the exact steps a team already takes, built the tool to mimic that, and tested its drafts against known-good human work before trusting it. That's the right order, and it's one you can borrow without a custom legal AI.
Try this
Pick one document-heavy review task your team does on deadline — contracts, applications, claims, vendor paperwork. Write down the actual steps a person takes, in order, including where they make judgment calls. That map tells you what could ever be a 'first pass' and what must stay with a person. Do the mapping before you even look at tools.
Who should care
Owners and operations leads at firms that do high-volume document review under deadline pressure — legal, accounting, insurance, procurement, compliance.
Skip this if
Your document work is low-volume, or you were hoping this means off-the-shelf legal AI is now safe to deliver to clients unreviewed — it isn't, and this firm says so plainly.
Sources: OpenAI — read the original