California's 2026 AI Ethics Guidance: What Changed From 2023 and Your 6-Point Action Plan
9/15/20263 min read
California's 2026 AI Ethics Guidance: What Changed From 2023 and Your 6-Point Action Plan
On May 14, 2026, the State Bar of California's Board of Trustees approved an updated Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law. Developed by the Committee on Professional Responsibility and Conduct (COPRAC), it replaces the original November 2023 version and, at the California Supreme Court's direction, becomes the first version to address agentic AI head-on.
If you practice in California and you use AI in any part of your workflow, this is the document that now describes how the State Bar expects your existing ethical duties to apply. Here is what changed, and the six things worth updating in your practice this quarter.
What the 2026 guidance is — and what it isn't
The Practical Guidance is exactly that: guidance. It does not create new rules. It maps duties you already owe — competence, confidentiality, communication, candor to the court, and supervision — onto the specific ways generative and agentic AI show up in legal work. The State Bar describes it as a living document that will be revised as the technology evolves.
Keep this distinction in mind: the guidance is advisory. A separate, still-pending effort would move some of these obligations into the enforceable text of the California Rules of Professional Conduct. That rule track is not yet adopted, and this article is about the guidance, not the proposed rules.
The biggest change: agentic AI
The 2023 version was written for generative tools that respond to prompts. The 2026 update responds to the California Supreme Court's August 22, 2025 directive to consider tools that can plan, select their own steps, and complete tasks with little or no human intervention — for example, revising pleadings across iterations, preparing discovery responses, coordinating document review, or handling client intake.
The core message on agentic AI is direct: autonomy does not satisfy a lawyer's duty to exercise independent professional judgment, and it does not alter any existing ethical obligation. A lawyer remains fully responsible for every output produced with AI assistance. The more a system can do on its own, the more deliberately you have to build in meaningful review and control.
What else moved
Beyond agentic AI, the update refines definitions and terminology and sharpens how the familiar duties translate to AI use:
Competence now clearly carries both a technical dimension (a reasonable understanding of how the tools work, including their tendency to produce confident but inaccurate output) and a judgment dimension (you cannot outsource the thinking). Confidentiality guidance remains firm that a lawyer should not input a client's confidential information into an AI tool without informed consent that accounts for the specific risks. Supervision is tied to the responsibilities lawyers hold under rules 5.1, 5.2, and 5.3, extended to how AI is configured and used across the firm. Communication points lawyers to consider, at the outset and as a matter develops, whether and how to tell clients that AI will be used — and to honor client instructions that limit or prohibit it.
Your 6-point action plan
Write or refresh a firm AI policy. Name approved tools, prohibited uses, and who signs off on agentic workflows. Then commit to updating it as tools gain autonomy.
Close the confidentiality gap. Identify which tools receive client data, confirm the data-handling terms, and build an informed-consent step before confidential information ever reaches an AI system.
Set a verification standard for anything court-facing. Require that every cited authority be independently confirmed before filing, regardless of which tool produced the draft.
Map your agentic exposure. List any tool that can act without a human pressing "go" at each step, and insert a review checkpoint before its work leaves the building.
Train, then re-train. Periodic training keeps the policy real. Tie it to the current guidance and to your own tools, not to a generic slide deck.
Decide your client-communication default. Choose a house position on AI disclosure and document it, so the decision isn't made ad hoc under deadline.
FAQ
Does the 2026 guidance ban any AI tool? No. It explains how your existing duties apply; it does not prohibit tools.
Is this the same as California's new AI rules? No. The guidance is advisory. A separate rule-amendment effort is still pending and not yet adopted.
Does it apply outside litigation? Yes. Competence, confidentiality, communication, and supervision apply across practice areas, not only to court filings.
If your firm's policy still reflects the 2023 landscape, this is the moment to align it. We help California firms turn the updated guidance into a concrete, defensible AI policy and workflow.