Is It Ethical for Attorneys to Use ChatGPT?

What the bar associations say

ABA Formal Opinion 512 (2023) addressed generative AI directly. It permits AI use for legal work, requiring attorneys to understand the data practices of any tool they use and comply with existing duties of confidentiality, competence, and supervision. Several state bars — including California, Florida, and New York — have issued parallel guidance. None prohibit AI. All require that client data be protected before submission to any AI platform.

The common thread across all state guidance: the question is not the tool, it is the data the tool receives. An attorney who submits a prompt containing no identifying client information has not made a disclosure under Rule 1.6. An attorney who submits a prompt containing a client's name, case number, or other identifying detail has.

The specific risk

The ethics concern is not using ChatGPT for legal analysis. It is submitting a prompt that contains identifying client information to an external AI server. That submission is a disclosure under ABA Model Rule 1.6 — one that most attorneys who use ChatGPT make without thinking about it, because the standard workflow of "write prompt, press enter" does not include a redaction step by default.

The failure mode is not malice. It is the frictionless nature of copy-paste: an attorney copies a passage from a client memo, pastes it into ChatGPT, and submits. The client's name was in the memo. The disclosure has occurred.

What counts as a reasonable precaution

A reasonable precaution is a measure that prevents client information from reaching the AI server at all. Naosuu provides this by intercepting prompts on the attorney's device, stripping identifying content before transmission, and restoring it after the response — entirely locally. No client data reaches OpenAI, Anthropic, or Google.

That is the mechanism the ethics obligations require a safeguard to address: the transmission itself. A policy saying "don't use client names in AI prompts" is not a precaution — it depends on the attorney remembering to comply under time pressure. An on-device technical measure that removes the transmission mechanically is.

Frequently asked questions

Does ABA Formal Opinion 512 prohibit AI use?

No. It permits AI use for legal work and identifies what attorneys must do to use it ethically: understand the tool's data practices, protect client confidentiality before submission, and supervise AI-assisted work product.

Is getting client consent a valid alternative?

It can be, depending on the jurisdiction and the sensitivity of the information. But client consent requires disclosing the AI tool's data practices in sufficient detail for the client to make an informed decision — and most clients do not understand what "submitting to OpenAI's servers" means. A technical safeguard that eliminates the exposure avoids the consent issue entirely.