How Law Firm Ropes & Gray Is Using AI to Reshape Transactional Due Diligence

While the legal industry is still debating whether AI will replace lawyers, top-tier law firm Ropes & Gray has already given its answer: let AI do due diligence, and let lawyers make judgments.

Freeing Lawyers from the Document Pile

Transactional due diligence is one of the most labor-intensive parts of law firm work. Behind a single M&A deal, there are often thousands of contracts, memoranda, and compliance documents that need to be reviewed page by page. Ropes & Gray is working with OpenAI to develop an AI system for transactional due diligence, and the core idea is not complicated: let AI first complete the preliminary processing of documents—grouping, summarization, and report generation—and then have lawyers review on that basis and provide legal advice to clients.

The division of labor in this process design is very clear. AI handles "quantity," and lawyers handle "quality." Document classification, key point extraction, and draft report generation—work that is highly repetitive but has relatively large tolerance for error—are handed to the model, while final judgment, risk warnings, and client consultation remain firmly in the hands of lawyers.

Why Due Diligence Is an Ideal Scenario for AI Deployment

There are many scenarios for deploying legal AI, but transactional due diligence has several natural advantages.

Documents are dense and relatively standardized in structure. Although due diligence involves many types of documents, contracts, agreements, disclosure letters, and similar materials all have mature formats and clause systems, making them easy for models to handle after targeted training.

Time pressure is real. Deal windows are often tight, due diligence cycles are compressed, and the marginal cost of manual review is extremely high. In this scenario, the efficiency gains brought by AI are easiest to quantify in value.

The human-in-the-loop design reduces risk. Ropes & Gray's solution does not let AI directly output conclusions; instead, it has AI produce intermediate materials for lawyer review. This means that when the model makes mistakes, professional judgment can serve as a backstop. This "AI first screening + lawyer review" architecture is currently the form of generative AI most widely accepted by the legal industry.

Human-in-the-Loop is a collaborative paradigm in AI system design that refers to introducing human review or intervention at key nodes in automated workflows rather than letting models make fully autonomous decisions. In legal scenarios, this design is particularly important: the model's output is not a final conclusion but intermediate material for lawyer review, and the lawyer's professional judgment constitutes the final line of defense. This architecture both leverages AI's efficiency advantages in document processing and preserves human understanding of risk, context, and client interests, so it is widely adopted in professional service fields with high fault-tolerance requirements.

The Signal Value of Law Firms Building Their Own AI

Ropes & Gray chose to develop in cooperation with OpenAI rather than directly purchasing off-the-shelf legal AI tools, and this choice itself is worth attention.

On the one hand, the business processes and knowledge systems of top-tier law firms are highly customized, and general-purpose tools struggle to fit their due diligence methodologies. Building in-house means the firm can encode its own review experience and internal standards into the system. On the other hand, this also shows that law firms are beginning to regard AI capabilities as part of their competitiveness rather than merely an outsourced tool.

From an industry perspective, attempts like this by leading law firms will create a demonstration effect. Once the process and results of AI due diligence are validated, small and medium-sized law firms will most likely follow suit, though the path may differ—they are more likely to purchase mature third-party products.

Legal AI tools can be broadly divided into two categories: general-purpose products (such as document review platforms based on large models) and customized systems (built in-house by law firms or enterprises). General-purpose tools deploy quickly and cost less, but they struggle to fit the due diligence methodologies and internal standards of specific law firms; customized systems require investment in data, computing power, and engineering resources, but they can encode review experience, risk preferences, and client confidentiality requirements into the process. Ropes & Gray's choice to build in cooperation with OpenAI is essentially treating AI capability as differentiated competitiveness rather than simply purchasing outsourced services. This path is more feasible for leading law firms, while small and medium-sized firms, constrained by resources, are more likely to adopt mature third-party products.

Questions Still Unanswered

Public information currently discloses only the process framework, and several key questions remain open. How accurate is the model when handling complex clauses, cross-references, and unusual clauses? How much rework by lawyers do AI-generated summaries and reports require? How are data security and client confidentiality obligations guaranteed in a self-built system? All of these determine whether this system can truly enter production environments.

What is certain is that the legal industry's path to AI adoption is becoming clearer: not replacement, but freeing lawyers from document processing so they can spend time on what truly requires professional judgment. Ropes & Gray's attempt is an early sample on this path and deserves continued observation.

Cross-references in contract documents refer to expressions in which clauses point to one another, such as "pursuant to Section 5.2" or "subject to Exhibit C." Such references require the reviewer to understand the interrelated logic of multiple clauses at the same time, posing a challenge to AI models: the model needs to track referential relationships across paragraphs and maintain consistency in summaries or reports. Unusual clauses (such as special arrangements that deviate from standard templates) also test the model's judgment, because they often follow no fixed pattern. Whether these complex situations can be accurately identified by the model directly affects the usability of AI due diligence systems in actual transactions.