ABA Formal Opinion 512, explained
In July 2024 the American Bar Association issued its first ethics opinion on generative AI. It did not create new rules. It applied the ones lawyers already live under to a new tool. Here is what it covers, in plain English.
Opinion 512 is best read as six existing duties, each viewed through the lens of a tool that can produce confident, well-formatted text that is simply not true. None of the six is new. What is new is that a lawyer now has to think about how each one applies the moment they open a chatbot.
1. Competence (Rule 1.1)
A lawyer must have a reasonable understanding of the AI tools they use, including their limits. The central limit is that generative AI can produce inaccurate or entirely fabricated output, so a lawyer cannot treat what it returns as reliable. Competence here means independent review of the output, not faith in it.
2. Confidentiality (Rule 1.6)
Before putting confidential client information into a generative-AI tool, a lawyer has to evaluate how that tool stores, uses, and retains what is entered. Some tools train on inputs or could expose them. For those, informed client consent may be required first, and reading the tool's terms and privacy posture is part of the duty.
3. Communication (Rules 1.2 and 1.4)
Depending on the circumstances, a lawyer may need to tell the client that generative AI is being used, particularly where it bears on the representation or where the client would reasonably expect to know. This is a judgment call, and it is one the opinion expects lawyers to make deliberately rather than by default.
4. Candor toward the tribunal (Rules 3.1, 3.3, 8.4)
Everything filed with a court has to be reviewed before it goes out. Submitting a fabricated citation, whether the lawyer knew it was fake or simply never checked, is the failure at the center of every AI-sanction case to date. The duty to verify what you file is the lawyer's, and using AI does not move it.
5. Supervisory responsibilities (Rules 5.1 and 5.3)
Managing lawyers have to put policies and training in place for how everyone at the firm, lawyers and nonlawyers alike, uses generative AI. An associate or paralegal pasting client facts into a chatbot is the firm's exposure, and supervision is the mechanism the rules expect to catch it.
6. Fees (Rule 1.5)
Fees must be reasonable and reflect time actually spent. A lawyer may not bill for hours that AI saved as if they were worked, and any AI costs charged to a client should be disclosed. Transparency about how AI factors into a bill is part of the fee duty.
What ties them together
Read together, the six duties describe a workflow, not a warning. Screen what you put in. Review what comes out. Tell the client where it matters. Supervise the team. Bill honestly. The hard part is not doing these things once. It is being able to show, months later, that you did them on a specific matter.
Candor is the record that shows you met each of these.
Screening before the model sees client data, a per-matter log of every AI use, verification prompts on the citations, and the attorney's sign-off, all on your firm's own model. Opinion 512 sets the duties. Candor produces the proof.
This is a reference summary that paraphrases a published ABA opinion; the authoritative text is the linked source, and this is not legal advice. Confirm the specifics against it, and note that your own state's rules and any court standing orders can add duties this summary does not capture. Candor is software, not a law firm.