Key takeaways
- The 2026 Practical Guidance replaces the November 2023 version and, at the California Supreme Court's request, addresses agentic AI[1].
- It names autonomously facilitating client intake as an agentic use and says supervision must match the system's access and autonomy[2].
- Do not put confidential client information into a tool that presents material confidentiality or security risks without informed client consent[3].
- Hourly bills must reflect time actually spent; flat fees may suit more predictable work[4].
What changed in 2026
The State Bar of California's Standing Committee on Professional Responsibility and Conduct published its first Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law in November 2023. The 2026 Practical Guidance replaces that version and, at the request of the California Supreme Court, addresses the challenges of agentic AI: systems that can plan, choose tools and carry out tasks without a human approving each step[1].
The guidance is not a new rule. It maps California's Rules of Professional Conduct onto AI use, duty by duty. For the country's largest legal market, with 22,821 offices of lawyers in 2023[5] employing 148,867 people[6], it is the practical reference most firms will be measured against.
Agentic AI, with intake as an example
The most practical change for intake teams is the guidance's treatment of agentic AI. Its examples include revising pleadings across multiple iterations, preparing discovery responses, coordinating document review and autonomously facilitating client intake. The guidance says the degree of lawyer diligence and supervision must correspond to the level of system access and autonomy[2].
In plain terms, the more a tool can do on its own, the more the firm must control what it may do. A tool that only drafts a note for a lawyer needs a review step. A tool that talks to prospective clients, collects their information and schedules consultations needs approved scripts, limits on what it can say, a clear route to a human and regular review of its conversations. The guidance also says lawyers must not let AI systems autonomously file documents, communicate with the court or make representations on the lawyer's behalf.
| Autonomy level | Example | Supervision that fits |
|---|---|---|
| Assistive | Summarizes a transcript for a lawyer | Lawyer reviews before use |
| Multistep | Drafts discovery responses from the file | Lawyer reviews every response and source |
| Client-facing | Answers inquiries and collects intake details | Approved script, topic limits, handoff rules, transcript review |
| Agentic with external actions | Schedules, sends messages or files on its own | Hard limits on actions; nothing to a court without lawyer approval |
Confidentiality: read the terms, not the marketing
As a general matter, a lawyer must not input any confidential client information into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent[3]. The guidance expects lawyers to understand how a product collects, uses, stores and discloses what users provide, and says reasonable efforts require more than relying on generalized marketing assurances: review the terms of use, privacy policies or vendor documentation.
- Check whether inputs are used to train models that serve other users.
- Confirm retention periods and whether the firm can delete data.
- Note that some products bind users to their terms simply by being used, so review before anyone tries a new tool with client data.
- Record the review in the firm's tool register with the date and the person who did it.
Telling clients
The guidance says lawyers must evaluate their communication obligations throughout a representation, based on the technology, the risks, the scope of the work and the client's sophistication. A lawyer must consider disclosing to the client the intent to use AI, including how it will be used and its benefits and risks; disclosure may be appropriate where AI use materially affects decision-making. AI use does not change the lawyer's obligation to make sure the client understands who is providing legal advice.
For intake, that last point is the one to design around. A prospective client talking to an AI tool should never be unsure whether a lawyer is involved. If an SI Employee such as Ava answers your California calls, the greeting your firm approves can say it is the firm's AI assistant, that it does not give legal advice and that the call is recorded, which California's all-party consent rules make essential.
Competence, candor and verification
The duty of candor to the tribunal cannot be delegated to AI, and the guidance says a lawyer must review all AI output used in submissions. The need is measurable. Stanford researchers found leading legal research tools produced incorrect information more than 17% of the time on their benchmark, one more than 34% of the time[7], and an earlier study found general-purpose chatbots hallucinated on 58% to 82% of legal queries[8].
Verification therefore belongs in the workflow, not in a policy binder: a named reviewer, a checklist for citations and quotations, and a rule that no AI-drafted document goes to a court or opposing counsel without a lawyer's approval.
Supervision and bias
Managerial and supervisory lawyers should set clear policies on AI use, including more autonomous tools, make reasonable efforts to ensure the people they supervise comply, and provide training on ethical and practical pitfalls. The guidance also flags the risk of biased output and says it can compound when several AI agents work together, with employment decisions and screening as examples. For intake, that means testing whether a tool treats callers differently by language, accent, age or matter type, and reviewing decisions to decline before they are final.
Billing
Hourly billing must reflect the time actually spent on the client's matter, recognizing that generative AI may reduce the time certain tasks take, and lawyers may consider alternative structures such as flat fees for more predictable assignments[4]. Put your approach in the fee agreement: what the firm bills when AI assists, whether any tool cost is passed on, and how flat fees are set. Our guide to written fee agreements has clause wording.
Other laws the guidance points to
The guidance reminds lawyers to stay reasonably informed about the laws and regulations that apply to the lawyer, the client and the specific use of the technology, not just the professional conduct rules. In California that list is long. Privacy law can apply to the firm itself: from January 1, 2025, a for-profit organization is a business under the CCPA if, among other tests, its annual gross revenue exceeds $26,625,000[9]. Recording law applies to AI that records calls, because California requires the consent of all parties to a confidential communication.
For client-facing tools, add a short legal check to the tool review: does it record, does it collect personal information covered by privacy law, and does the privacy notice on the firm's website describe it accurately? Our article on California call recording rules for intake covers the recording side.
A compliance checklist for California firms
Inventory
List every AI tool and AI feature in use, including features built into email, document and meeting software.
Classify by autonomy
Mark each as assistive, multistep, client-facing or agentic, and set the supervision level from the table above.
Review data terms
Read terms, privacy policies and vendor documentation; record whether client information may be used and on what conditions.
Decide disclosure
Add an engagement letter paragraph and decide when a matter-level conversation is needed.
Set review rules
Name reviewers and checklists for anything filed, sent to opposing counsel or said to a prospective client.
Train and test
Train staff, run bias and accuracy tests on client-facing tools, and record both.
Revisit
Review the register and the tools at least twice a year; the guidance itself was rewritten within three years.
Example: an intake assistant in a Los Angeles firm
Example: a fictional twelve-lawyer employment firm in Los Angeles adds an AI intake assistant to its phone line. Treating it as client-facing under the 2026 guidance, the firm approves every question, blocks discussion of claim value or deadlines, and routes any caller who asks what they should do to a callback from an attorney. The intake manager reviews twenty conversations a week and compares acceptance decisions across languages. The vendor confirms in writing that recordings and transcripts are not used to train models for other customers. The engagement letter for accepted clients explains how the firm uses AI tools, and the billing policy states that only actual time is billed.
For the national baseline, see ABA Formal Opinion 512; for how other states compare, see New York's guidance and Texas Opinion 705.
Frequently asked questions
What is the State Bar of California's guidance on generative AI?
The Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law. The 2026 version replaces the November 2023 version and adds guidance on agentic AI at the California Supreme Court's request.
Does the California guidance cover AI intake tools?
Yes. It names autonomously facilitating client intake as an example of agentic AI and says supervision must correspond to the system's access and autonomy.
Do California lawyers have to tell clients they use AI?
The guidance says lawyers must consider disclosing intended AI use, how it will be used and its benefits and risks, and that disclosure may be appropriate where AI materially affects decision-making.
Can California lawyers bill for time AI saved?
No. Hourly billing must reflect time actually spent. The guidance suggests flat fees as an option for more predictable work.
References
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
U.S. Census Bureau (2023). County Business Patterns 2023: state file (cbp23st) (opens in a new tab) โ
U.S. Census Bureau (2023). County Business Patterns 2023: state file (cbp23st) (opens in a new tab) โ
Stanford Institute for Human-Centered Artificial Intelligence (HAI) (2024). AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries (opens in a new tab) โ
Stanford Institute for Human-Centered Artificial Intelligence (HAI) (2024). AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries (opens in a new tab) โ
California Privacy Protection Agency (2025). Updated Monetary Thresholds in CCPA (opens in a new tab) โ