What does ABA Formal Opinion 512 require firms to disclose about AI use?
The opinion applies the rules a firm already lives under to generative AI. Disclosure turns on your jurisdiction; accountability doesn't.
Answered in short
5 things that decide this
- 01ABA Formal Opinion 512, issued July 29, 2024, is the ABA's first ethics opinion on generative AI. It applies the existing conduct rules to AI rather than creating a new, standalone disclosure rule.
- 02Whether a firm must tell clients about AI use depends on where you practice. As of June 2026, 15 US jurisdictions have formal AI ethics opinions and 38 have some form of guidance, and they don't all say the same thing.
- 03The one constant across every opinion: AI may assist, but the lawyer remains accountable for every output.
- 04What binds a firm is its own state's rules, not the ABA opinion itself. California's proposed rule changes of March 2026 carry disciplinary authority, a sign that guidance is turning into hard rules.
- 05The durable answer is a supervision record mapped to each state you practice in: where AI assisted, who reviewed the output, and what the client was told.
What Opinion 512 changed, and what it didn't
Opinion 512 changed the baseline, not the rulebook. Since July 2024, no US firm can call generative AI an open question: the ABA has said how the profession's existing duties apply to it. It did not hand firms a single disclosure rule, though. ABA opinions guide; state rules bind, and states move at different speeds.
The state layer is where disclosure gets decided. Fifteen jurisdictions had issued formal ethics opinions on AI and 38 had published some form of guidance as of June 2026. A firm practicing across three states can face three different answers to "must we tell the client".
California shows the direction of travel. Its proposed rule changes of March 2026 carry real discipline behind them, which moves AI use from a guidance conversation to a discipline conversation.
How a firm handles disclosure when the answer varies
You handle the disclosure question by making supervision provable rather than betting on one reading of one opinion. Map a disclosure position to each state you practice in. Add a required human check before AI-assisted work leaves the firm. Then log each matter's AI use against the person who reviewed it. Any bar inquiry gets a record, not a recollection.
Mapping matters because the opinions genuinely differ. A blanket "we disclose everything" policy over-discloses in some states and still under-documents in all of them. A policy mapped to each state's actual guidance is both lighter and stronger.
This is a build problem as much as a policy problem. We build legal AI where visibility is structural. LexPair, a legal lead-generation platform Hashlogics built, matches each case to the right practice area with AI-assisted logic. Its admins keep full visibility into every lead. That visibility discipline is what a managing partner wants over drafting and research tools too.
- AI assistsA draft, a summary or research is produced.
- CheckpointA lawyer reviews before anything leaves the firm.
- Disclosure decisionClient told per your state's rule.
- LogMatter, tool, reviewer, date.
- On requestThe record a bar inquiry sees.
Accountability is the constant. Disclosure is the variable you map per state.
Related questions
01Does Opinion 512 itself bind our firm?+
No. ABA formal opinions guide rather than bind; what binds you is your own state's conduct rules and ethics opinions. Where your state has adopted or echoed the ABA's reasoning, ignoring it is a bad bet.
02Do we have to tell every client we use AI?+
Not everywhere, and not for every use. The answer depends on your state's guidance and on what the AI actually touches in the matter. A workable policy names the uses, maps each state you practice in, and errs toward telling the client where the guidance is unclear.
03What should the supervision record contain?+
Enough that an outsider could rebuild the story of the work. Record which matter, which tool, what it produced, who reviewed it, what changed after review, and what the client was told. A record like that answers most disclosure questions before they're asked.
Related
- Software for law firms →Our legal hub: what firms actually buy.
- AI agent development →Our service page: legal AI built with the checkpoint and the audit trail in place.
- What happens when a lawyer cites an AI-hallucinated case? →The failure mode these opinions exist to prevent.
- What does the SRA's AI misuse warning notice require? →The UK version of the same question, with enforcement behind it.
- Is legal AI safe for client confidentiality? →The confidentiality half of the ethics question.
