Test a sample of chatbot transcripts for client information and wrong law
Key facts
- Human involvement
- YellowClose supervision
- How AI is used
- On a closed sample the lawyer selects, flag turns that look like legal advice, citations, personal data or a promise to file or contact a court.
- What the lawyer must review
- You read every flagged turn and a random slice of the rest. Opinion 512: uncritical reliance on generated output can breach competence, and client information must not be put into a tool that discloses it. California's guidance: competence is more than deleting hallucinations, and candour to a tribunal cannot be delegated if a transcript is later used in court. NYC Bar Formal Opinion 2024-5 covers the same confidentiality, competence and supervision duties.
- Written for / version
- United States versionUS
Who does what
| Role | What they do |
|---|---|
| Assistance | Potential support includes identifying points for review. Confirm the tool’s capability and permitted scope. |
| Lawyer | Set the scope, supervise the work closely, verify every part relied on, resolve omissions and exceptions, and approve the exact output. |
Checklist
- Confirm that the actual tool, feature and processing terms permit use of this information, including access, retention, training and disclosure conditions.
- Confirm the output against the client’s instructions and the complete relevant source material, including omissions or contradictions.
- Resolve exceptions and record the responsible lawyer’s approval of the exact output and intended use.
Forms and precedents for this task
| Form or precedent | Kind | AI use | Summary |
|---|---|---|---|
| Note escalating a client-facing tool transcript to a lawyer (United States) | Memo | AI use Yellow | An internal note that lifts a transcript out of a client-facing tool and records why a lawyer must take the next step. |
Signed-in accounts can read the full text of each one. All precedents.