Where AI Fits In Due Diligence

It is 9:30 p.m., the data room is still moving, and the deal team wants a clean view of what changed since this morning. In that moment, AI earns its keep as a throughput tool. You can point it at a defined document set and get an AI risk flag list, an AI clause summary per agreement, and an AI extraction table of terms like change of control, assignment, termination, exclusivity, and governing law, fast enough to keep pace with the clock. What does not change is the professional posture. The AI outputs are drafts for attorney judgment, and anything that becomes a diligence finding still needs attorney sign-off before it goes to the client or the deal team.

When you use AI well in diligence, you are not asking for a magical read of 800 contracts. You are setting up a workflow where AI takes the first pass at organizing the pile, clustering similar agreements, surfacing outliers, and producing a consistent starting format for review. That is how you get leverage without losing control of the review. The work still looks like diligence. You are deciding what matters for this transaction, what is missing, what needs escalation to a specialist, and what can be reported with confidence after you have verified it to the right depth for the risk tier.

A typical first screen you will see is a dashboard view. It usually includes counts, categories, and confidence scores that look precise enough to forward. The skill you are building in this course is knowing what parts of that dashboard are a helpful routing layer today, and what parts only become usable after verification and sign-off. Take a look at the kind of dashboard output we mean.

Professional responsibility in AI-assisted review

In diligence, the professional-responsibility questions show up in ordinary places. Someone asks you to paste a “representative contract” into a tool to get a quick summary. A teammate wants to rely on the AI risk flags to decide what to read. The vendor offers a feature that stores your prompts to improve its model. These are workflow decisions, not theory, and they map to three plain obligations you already live with.

• Competence means you must understand what the tool is doing well enough to use it appropriately and to catch when it is outside its lane. If you cannot explain why a clause was flagged, what document set was analyzed, and what the output does not cover, you cannot responsibly rely on it in a live deal.
• Confidentiality means you protect client information and privileged material when you use AI. The consequence of getting this wrong is not embarrassment. It can be privilege exposure, client harm, and an obligation to disclose what happened.
• Duty to supervise means you remain accountable for the work product, even if AI did the first pass. You need a review structure where junior lawyers, staff, and tools are all producing outputs that are checked to the right standard before anything is communicated as a finding.

Callout: Confidentiality boundary is a tool choice
Do not put privileged or confidential deal material into non-enterprise AI tools. Use your organization’s approved platform and data-governance controls, and confirm retention, access controls, and who can see prompts and outputs before you upload or paste any client content.

You will often be choosing between two setups that can produce similar-looking summaries, but with very different confidentiality and supervision implications. The right comparison is not “which summary reads better.” It is which setup gives you control over what leaves the deal environment, how it is stored, and how you can document the review and sign-off process. Put those two setups side by side.