AI Ethics Opinions Tracker
Every United States bar association ethics opinion, official guidance document, task force report, and court rule on artificial intelligence in legal practice — labeled by instrument type, because weight and citability differ, and sorted by issuance date so the newest authority is always on top.
Awaiting first update · Reviewed quarterly and on each new instrument
Add an Instrument
Passcode-protected editor for adding a newly issued ethics opinion, guidance document, report, or court rule.
The Agentic Frontier
Only two bar instruments in the country squarely treat autonomous AI agents as a distinct category. The ABA has issued nothing formal on agentic AI.
Formal Opinion 2026-208, Chatbots and AI Agent Communications
The first U.S. state bar ethics opinion to analyze autonomous AI agents as a distinct category. Requires bot-identifies-as-bot disclaimers, prompt lawyer notification and review, conflict screening at intake, and a lawyer-approval gate on autonomous action. Casts the Rule 5.3 duty as supervision of the vendor, not of the agent.
COPRAC 2026 Practical Guidance
Revised at the California Supreme Court's direction to address agentic AI. Central principle: the greater the autonomy afforded an AI system, the more important supervisory controls and verification become. Parallel proposed rule amendments would make these duties binding if adopted.
From the Blog: Securing the Autonomous Agent
Once you know what the bar instruments say, the next question is how to operationalize them inside a firm. Our article translates the ethics layer into a five-stage governance framework — inventory, contain, localize, supervise, and disclose — with source links to OWASP, the ABA, and the sanctions docket.
Read Securing the Autonomous AgentOhio flag: Ohio has not issued a formal ethics opinion on AI. Its only lawyer-facing pronouncement is the Ohio Board of Professional Conduct's Ethics Guide (April 2026), expressly nonbinding staff guidance that does not reflect the views of the Board, its Commissioners, or the Supreme Court of Ohio.
Instruments by Quarter
Instrument Types
Where Everyone Agrees
- 01Use is permitted. No opinion in the country bans generative AI outright.
- 02The lawyer remains responsible. Ultimate responsibility for work product never shifts to the tool.
- 03Verification is mandatory. Every opinion requires independent verification of AI output, especially citations.
- 04Competence and confidentiality govern. Duties are grounded in Rule 1.1 Comment 8 and Rule 1.6.
Where They Split
- Client disclosure and consent
The sharpest divide. West Virginia requires informed consent confirmed in writing for any generative AI use. Florida and Texas require consent only where confidential information reaches a third-party tool. New York City and Illinois expressly decline any blanket duty. North Carolina sits in the middle.
- Billing
North Carolina and the ABA are most explicit that AI-saved time may not be billed as attorney time. North Carolina permits an AI subscription as a pass-through cost.
- Embedded vs. standalone tools
New York City expressly exempts embedded, routine tools such as Word, Westlaw, and Lexis from disclosure. Most opinions are silent.
- Agentic autonomy
Only Oregon and California treat autonomous AI agents as a distinct category. The ABA has issued nothing formal on agentic AI.
| Date | Jurisdiction | Instrument | Type | Disclosure Posture | Source |
|---|---|---|---|---|---|
| No entries match those filters. | |||||
Methodology and Counting
This database uses a narrow, type-labeled approach. A numbered formal ethics opinion is recorded as such; a task force report, bar-counsel article, court rule, or judicial-branch policy is labeled by its own character, because the weight and citability of each differ. Roughly 15 U.S. jurisdictions plus the ABA have issued numbered formal ethics opinions on generative AI, and about 38 have issued something official once guidance, reports, and court rules are counted.
Vendor pages claiming "25+," "30+," or "35+ states" are only accurate under a broad definition that folds in reports, FAQs, court policies, and bar-counsel articles. Those figures are not wrong so much as differently scoped.
Entries recording that no instrument was identified are deliberate. An absence of located material is not the same as an affirmative statement by the body, and where a negative finding rests on a single tracker it is flagged in that entry's note.
The judicial-branch layer dwarfs the bar-opinion layer: standing orders, local rules, and sanctions decisions number in the hundreds. Check the assigned judge's standing order before every filing. See also the AI Sanctions Tracker.
Compendium currency date: August 18, 2026. Reviewed quarterly; new instruments are added as they issue.