Opinion 512, rule by rule
| Model Rule | What the opinion says, in short | Practical step |
|---|---|---|
| 1.1 Competence | Understand the tool’s capabilities and limits, including that it can produce inaccurate output; keep up as it changes. | Train everyone who uses it; verify output. |
| 1.6 Confidentiality | Assess the risk that information will be disclosed or accessed by others. For self-learning tools, informed consent is generally needed before inputting information relating to the representation; boilerplate in an engagement letter isn’t enough. | Know where each tool sends data; get real consent where required. |
| 1.4 Communication | Tell clients about AI use when it affects the representation or they ask. | Say how you use AI in plain terms. |
| 3.1, 3.3, 8.4(c) Candor | Check AI output, especially citations, before relying on it with a tribunal. | Open and read every cited authority. |
| 5.1, 5.3 Supervision | Managers set clear policies and train lawyers and nonlawyers; vet vendors. | A written AI policy; vendor due diligence. |
| 1.5 Fees | Hourly billing covers time actually spent; don’t charge clients for learning a tool you’ll use generally; costs must be reasonable. | Review billing practices for AI-assisted work. |
One-page checklist for small firms
- List the AI tools in useIncluding features inside software you already have.
- For each, find where data goesWhich company processes it, whether it trains on it, how long it’s kept.
- Decide what may go into each toolAnd what never does.
- Get informed consent where requiredSpecific to the use, not boilerplate.
- Verify before relyingFacts, figures and every citation.
- Write a short firm policyWho may use what, for which tasks.
- Train lawyers and staffIncluding the tools’ limits.
- Review billingHourly time reflects time actually spent.
- Revisit yearlyTools and guidance change.
Does self-hosted AI change the confidentiality analysis?
The opinion’s consent concern centers on tools that may disclose or use information beyond the lawyer’s control, including self-learning tools that could surface one client’s information to others. A model that runs on a computer in your office, sends nothing to an AI provider and isn’t training a shared model on your matters changes the facts that analysis turns on. Whether consent is still needed for a particular use is your judgment under your own state’s rules.
Several state bars have issued their own guidance, including the Florida Bar (Ethics Opinion 24-1) and the State Bar of California (practical guidance on generative AI). ABA opinions are advisory; your state’s rules and opinions govern.
Questions
Can lawyers use ChatGPT?
Yes, consistent with Opinion 512’s duties. The confidentiality question depends on the plan and whether client information goes in.
Do I need client consent to use AI?
Opinion 512 says informed consent is generally required before inputting information relating to the representation into a self-learning tool. Check your state’s rules.
Can I bill clients for time saved by AI?
Under Opinion 512, hourly billing must reflect time actually spent, and lawyers shouldn’t charge for learning a tool they’ll use generally. Flat fees must still be reasonable.
Is Opinion 512 binding?
No. ABA formal opinions interpret the Model Rules and are advisory. State rules and state bar opinions govern.
When was Opinion 512 issued?
July 29, 2024, by the ABA Standing Committee on Ethics and Professional Responsibility.