AI ethics guide · New York
Can a New York lawyer put client documents into an AI tool?
What NYC Bar Formal Opinion 2024-5 says about client confidentiality, and the other duties that come with using AI.
Not legal advice. This is a plain-English summary of NYC Bar Formal Opinion 2024-5, written by the team that makes ClientSafeAI, and it hasn't been reviewed by a lawyer. Read the source itself, and ask the New York bar's ethics hotline if you're unsure how it applies to you.
| Document | NYC Bar Formal Opinion 2024-5, Generative AI in the Practice of Law |
|---|---|
| Issued by | New York City Bar Association Committee on Professional Ethics |
| Date | August 7, 2024 |
| Status | Advisory bar association opinion |
The short answer
Only with consent if the tool shares what you put in. The NYC Bar’s Formal Opinion 2024-5 says that without client consent, a lawyer must not put confidential client information into any generative AI system that will share it with third parties. Even with consent, you should avoid details that identify the client. And a tool that doesn’t share your input still has to have adequate security.
What it says about confidentiality
Opinion 2024-5 is laid out as a table of duties and guidelines, and confidentiality comes first. It starts from how these systems work: they can use prompts, uploaded documents and other inputs to train the AI, and may share that input with third parties or use it for other purposes. Systems that share input with third parties are sometimes called “open” systems.
The rule that follows is direct. Without client consent, a lawyer must not input confidential client information into any generative AI system that will share it with third parties. Even with consent, the lawyer should avoid entering details that could identify the client. Consent isn’t needed when no confidential information goes in at all, for example because it has been anonymized. Consent also has to be informed: the client must understand what sharing the information could lead to.
“Closed” systems, which keep input within the firm’s own protected environment, aren’t a free pass. Even a system that neither trains on nor shares input may lack reasonable security. And inside a closed system, the firm still has to stop confidential information reaching people at the firm, or other clients of the firm, who shouldn’t have it. Where a system uses client information, the firm must make sure it applies any ethical screens the conflict rules require.
The other duties it covers
| Supervision | A lawyer using a generative AI chatbot for client intake must adequately supervise it. A chatbot might fail to say it isn’t a lawyer, or appear to give legal advice. |
|---|---|
| Competence | Output may be false, inaccurate or biased. Treat it as a starting point and review it critically. |
| Candor | Check that cited decisions are genuine and properly cited. The opinion points to Mata v. Avianca and United States v. Cohen, both in the Southern District of New York. |
| Other rules | The opinion also covers fees, advertising and solicitation (Rules 7.1 and 7.3) and compliance with the law, including privacy and cybersecurity law (Rules 1.2(d) and 8.4). |
In practice
- Read each tool’s terms to see whether it shares or trains on input. Don’t go by the marketing.
- Keep confidential client information out of any tool that shares input with third parties unless the client has given informed consent.
- Where you can, anonymize before you prompt. If no confidential information goes in, consent isn’t needed.
- If the firm runs an intake chatbot, assign someone to supervise it.
Worth watching
The New York State Bar Association’s April 2024 Task Force report is a separate, longer document that suggests adopting AI-specific rules. The NYC Bar opinion considers new rules premature and applies the existing ones instead.
How this compares with ABA Formal Opinion 512
The ABA's national opinion requires a client's informed consent before information relating to the representation goes into a self-learning AI tool, and says boilerplate consent in an engagement letter isn't enough. Read the Opinion 512 summary and firm checklist.
Keep client documents on a server that’s only yours
A Private AI Workspace runs on a GPU server ordered for one firm. Your documents, the AI model and the database stay on it, it never trains on your files, and answers cite the page they came from. It costs $499 a month, flat.
Sources
- NYC Bar — Formal Opinion 2024-5, Generative AI in the Practice of Law (PDF)
- NYSBA — Report and Recommendations of the Task Force on Artificial Intelligence, April 2024 (PDF)
Spotted something out of date? Tell us and we'll fix it.
Not legal advice. This is a plain-English summary of NYC Bar Formal Opinion 2024-5, written by the team that makes ClientSafeAI, and it hasn't been reviewed by a lawyer. Read the source itself, and ask the New York bar's ethics hotline if you're unsure how it applies to you.