Many legal organizations have responded to generative AI by developing policies. That is an important step. Policies can establish approved tools, identify prohibited uses, and communicate expectations about confidentiality and verification.
But a policy cannot supervise work. It cannot determine whether a particular use was appropriate, whether the person using the tool understood its limitations, or whether the resulting work was reviewed with the right level of care.
Responsible AI use is therefore not simply a technology or compliance issue. It is a management practice.
The finished product does not tell the whole story.
A polished memorandum may conceal fabricated authorities, incomplete research, subtle analytical errors, or confidential information entered into an unapproved system. Conversely, responsible use of an approved tool may improve organization or help a lawyer identify questions that deserve further research.
Supervisors cannot evaluate those differences by looking only at the final document. They need enough information about the process to assess the work responsibly.
Four questions should become routine.
1. What tool was used?
Different systems have different terms, safeguards, and capabilities. Before considering the quality of the output, determine whether the tool was approved for the work and information involved.
2. What information was provided?
The prompt matters. So do any documents or facts entered into the system. Lawyers must consider confidentiality, privacy, privilege, security, and organizational policy before providing information to an AI tool.
3. What role did AI play?
“I used AI” can describe very different activities: brainstorming issues, summarizing a document, generating search terms, conducting research, organizing an argument, drafting text, or revising prose. The nature of the task should influence the type and intensity of review.
4. How was the output verified?
Verification must be more than asking the same tool whether its answer is correct. Authorities, quotations, facts, reasoning, and completeness should be checked against reliable sources. The lawyer should also be able to explain the analysis independently of the system that helped produce it.
Accountability cannot be delegated to a tool.
AI may assist with legal work, but professional responsibility remains with the lawyers and organizations using it. Junior lawyers need clear expectations and a safe way to ask questions. Supervisors need sufficient understanding of both the technology and the assignment to provide meaningful oversight.
This requires more than warning people about hallucinations. It requires organizations to teach good process: when AI use is appropriate, how it should be disclosed, what verification looks like, and when a question must be escalated.
One practice to try
Add a short “process conversation” to the review of AI-assisted work. Ask what tool was used, what information it received, what task it performed, and how the result was verified. The goal is not surveillance. It is to make sound judgment and responsible supervision part of the normal workflow.
Organizations that treat AI primarily as a policy problem may miss the more consequential challenge. The quality of AI-assisted legal work will depend on the quality of the management surrounding it.
— Arlety Campos Bowman and Regan Farrar Cucinell