California AI Ethics Rules for Lawyers

The official instruments

The first mover, about to move again

California opened the entire field: COPRAC’s Practical Guidance, approved November 16, 2023, was the first official state instrument on lawyers and generative AI, and its influence shows everywhere; New York City’s Formal Opinion 2024-5 expressly modeled itself on the California chart. The format is distinctive: rather than a narrative opinion, the guidance maps each Rule of Professional Conduct to concrete dos and don’ts, which makes it unusually usable as a firm-policy source document.

The substance anticipated what nearly every later instrument adopted. Confidentiality: do not input confidential client information into a tool that lacks adequate security and confidentiality protections; anonymize where possible; consider consent where risk remains. Competence and diligence: understand the tool’s risks and limitations before use, and review every output; AI does not get the benefit of the doubt. Billing: a lawyer may bill for time actually spent using and reviewing AI, never for the time AI saved, and fee agreements should address AI charges. Supervision: firms must train and supervise lawyers and staff on permissible AI use. Candor: verify before filing. The 2026 revision, prepared at the California Supreme Court’s request, carries the same architecture forward and extends it to agentic tools that act in sequences rather than single responses, the first official state instrument to do so.

The pending rules are the national story

What makes California the jurisdiction to watch in 2026 is procedural, not substantive. On August 22, 2025, the California Supreme Court directed the State Bar to consider converting guidance into binding rules. COPRAC responded with proposed AI-specific amendments to six rules: 1.1 (competence, with AI-output verification written into the rule), 1.4 (communication about AI use), 1.6 (confidentiality), 3.3 (candor, aimed at AI-fabricated citations), and 5.1/5.3 (firm-leader policies and nonlawyer supervision). The committee approved the package for public comment on March 13, 2026; the 45-day window closed May 4, 2026; Board of Trustees and Supreme Court action remained pending as of this page’s last verification.

The significance is hard to overstate: every other jurisdiction has so far applied existing rules to AI through opinions and guidance. California would be the first to make AI duties disciplinary text. For multi-state firms, that converts “best practice” into “rule” in the country’s largest legal market, and the verification duty proposed for Rule 1.1 would be the strongest anti-hallucination mandate in the country. The tracker lists the proposal with status “proposed” and will move it the week it changes; the changelog and alerts exist for exactly this entry.

What California lawyers should do now

Build to the proposed rules, not just the guidance, because the delta is small and the direction is set. Concretely: an approved-tools register with documented security and confidentiality terms (the guidance’s “adequate protections” test); anonymization as default practice with consent for residual risk; a verification workflow treating the proposed Rule 1.1 standard as already operative; engagement-letter AI language to satisfy the communication duty; billing limited to actual time with AI costs addressed in the fee agreement; and firm training with acknowledgment, since the proposed 5.1 amendment targets firm-leader policies specifically. The policy template implements each element; the comparison table places California’s consent posture among the states.

One layer the guidance does not cover: courts. California has no statewide court AI rule, so the obligations that exist are judge-by-judge standing orders, particularly in the federal districts. The court orders guide covers the per-filing check that California practice still requires.