California AI Ethics Rules for Lawyers: What Binds You Now, and What Is Coming

As of September 2026, a California lawyer using AI is governed by the Rules of Professional Conduct already on the books, read through the State Bar's Practical Guidance, which was revised May 14, 2026 and is advisory rather than binding. Six AI-specific rule amendments would convert those duties into enforceable text, but they remain proposals awaiting approval.

That two-track structure is the whole story, and confusing the tracks is the mistake to avoid. One instrument can inform a disciplinary charge; the other would be the charge.

This article is for general information and is not legal advice. Confirm the current status of any rule before relying on it.

Two instruments, two very different weights

Amendments take effect in California only when the Supreme Court approves them, the path rule 8.3 took in 2023. Nothing in the AI package has reached that stage. Treat any source saying California "has adopted AI rules" as wrong, and treat the proposals as a preview of how a disciplinary authority already reads the existing ones.

Both tracks trace to an August 22, 2025 letter from the California Supreme Court directing the State Bar to move the guidance principles into the rules and address agentic AI. That is why California sits ahead of every jurisdiction in our 50-state survey: elsewhere the output is opinions, here it is rule text.

The duty map

One quirk is worth knowing. Rule 1.1(a) prohibits failing to perform competently "intentionally, recklessly, with gross negligence, or repeatedly," so ordinary negligence is not a disciplinary violation here, a higher threshold than the ABA Model Rule. The guidance leans on the affirmative framing instead: the lawyer, "and not the technology," retains "control over strategic decision-making and the ultimate exercise of professional judgment."

Confidentiality is the sharp edge

Everything else on that list is a workflow habit. This one is a tool-selection decision you make once and live with.

The Practical Guidance states it directly: a lawyer "must not input any confidential information of the client into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent." It names the mechanism, that products "often utilize the information that is input by the user" to "further train or refine the AI," and closes the escape hatch: "Reasonable efforts require more than reliance on generalized marketing assurances."

The proposed Rule 1.6 comment goes further, redefining the operative verb. Under the draft, "reveal" would include "exposing confidential information to technological systems, including artificial intelligence tools, where such exposure creates a substantial risk that the information may be accessed, retained, or used" inconsistently with the duty, with "whether data is retained or used for model training" a factor the lawyer must weigh. Pasting a client's facts into the wrong prompt box would not be a step toward a disclosure. It would be the disclosure.

That makes tool selection a procurement question, not a prompting question: before client information goes in, know where it lands, who can read it, whether it trains a model, and how long it is kept. We work through that in can I upload client documents to ChatGPT, and the privilege half in does AI waive attorney-client privilege. It is also why CaseRead isolates each firm's documents in its own database schema rather than filtering rows in a shared table.

The 2026 revision extends this to agentic systems with standing access to email, document management, or client files: such a system may "unintentionally disclose confidential information (including across different matters)," and must never permit "autonomous external transmission of client information" without human review.

Billing: you cannot charge for time you did not spend

A lawyer "may charge for actual time spent," meaning time crafting prompts and reviewing outputs. But "hourly billing must reflect the time actually spent on the client's matter, recognizing that generative AI may reduce the time required for certain tasks." If AI turns a six-hour memo into ninety minutes, you bill ninety minutes.

The guidance offers a constructive answer rather than a pay cut: use flat fees for predictable assignments and build the savings into your rate. Whatever the structure, the fee agreement should state how fees are calculated and "the extent to which generative AI factors into the lawyer's rates."

On costs the line is clean. General AI subscriptions are overhead, "similar to library maintenance or general computer systems," and belong in your fee, not on the invoice. Per-use fees for specialized tools built for one matter may be billed, but only with fee-agreement disclosure, at actual cost, and with "no markup or profit element" absent informed written consent. Full treatment in billing for AI-assisted work. Note what the amendment package omits: Rule 1.5. This analysis is advisory only.

Disclosure: required, recommended, and neither

To your client. No flat requirement. A lawyer "must consider disclosure to their client that they intend to use AI," and it "may be appropriate where AI use materially affects decision-making processes." One hard obligation does exist: you "must also follow any applicable client instructions or guidelines that may restrict or limit the use of AI." Outside counsel guidelines saying no AI bind you whatever the rules say.

To the court. No statewide requirement for lawyers. But the guidance instructs you to "check for any rules, orders, or other requirements in the relevant jurisdiction that may necessitate the disclosure of the use of AI tools" and to comply. In California these arrive through individual judges' standing orders, not uniform even within one courthouse. The Northern District publishes standing orders for all its judges in one place; most California courts have an equivalent page. Read your judge's before filing. The national picture is in do lawyers have to disclose AI use to courts.

Neither. California's courts have binding AI rules that point inward: Rule of Court 10.430, effective September 1, 2025, required every court not prohibiting generative AI to adopt a use policy by December 15, 2025. It governs court staff and judges, not practitioners.

Where California diverges from ABA Formal Opinion 512

ABA Formal Opinion 512, issued July 29, 2024, maps the same duties, but it interprets Model Rules that bind no one. California interprets rules binding every licensee, and the amendments would put AI language inside those rules.

California also reaches autonomy by name: lawyers must never let AI "autonomously file documents, communicate with the court, or make representations on the lawyer's behalf." And its scope is broad by design, covering "any current or future computational system that generates outputs, recommendations, or actions," including research platforms incorporating AI "even when not traditionally thought of as AI." These duties do not wait for you to open a chatbot. Compare our Utah deep dive, where the operative authority is a sanctions decision rather than draft rule text.

A compliance checklist for a California solo or small firm

1. Pick tools on retention terms, not features. Confirm in the terms of use whether inputs train the model, who can access them, and how long they are kept.
2. Write the policy down. Proposed Rule 5.1 would require managerial lawyers to establish internal policies governing AI use. Two pages naming approved tools, prohibited inputs, and the verification step will do; our law firm AI policy template is a starting draft.
3. Verify before filing, every time. Proposed Rule 3.3 Comment [3] would require confirming that no cited authority "is fabricated, misstated, or taken out of context."
4. Address AI in the engagement letter. How fees are calculated, whether AI costs may be billed separately, and whether the client restricts AI. One paragraph covers Rules 1.4 and 1.5.
5. Supervise the humans too. Rule 5.3 reaches paralegals and contractors: same tool list, same prohibited inputs, same training.

Verification is the one duty you cannot delegate

Every California instrument, binding and advisory alike, converges on one act: a human confirming the authority cited is real, current, and says what the brief claims. Rule 3.3 Comment [2] already treats citing an overruled decision or repealed statute as a false statement of law, whether a model wrote it or you did. The method is in how to verify AI-generated citations.

That is the problem CaseRead is built around: it searches your firm's case files and the public law together, scoped to the matter you are on, and every answer traces to a source the system actually retrieved. What it cannot verify, it flags. If a draft came out of a general-purpose tool, run it through the free Hallucination Shield first: each citation it finds, up to 50 per run, is checked for existence and support.

Frequently asked questions

Can California lawyers use ChatGPT for legal work?

Yes. Neither the State Bar's Practical Guidance nor the proposed rule amendments prohibit generative AI. The constraints are conditions of use: understand the tool before deploying it, keep confidential client information out of any system presenting material confidentiality risks absent informed client consent, verify every citation before it reaches a court, and bill only time actually spent.

Are California's AI ethics rules for lawyers binding yet?

Not as AI-specific rule text. As of September 2026 the amendments to Rules 1.1, 1.4, 1.6, 3.3, 5.1, and 5.3 are proposals that completed a second public comment period on August 6, 2026. They still require Board of Trustees approval and California Supreme Court adoption. The underlying rules themselves already bind you, and always have.

Does California require lawyers to tell clients they use AI?

There is no flat disclosure requirement. The 2026 Practical Guidance says a lawyer must consider disclosure, and that it may be appropriate where AI use materially affects decision-making. A lawyer must also follow client instructions restricting AI use. The proposed Rule 1.4 comment would fold that evaluation into the duty to keep a client reasonably informed.

Can California lawyers bill for time that AI saved?

No. The Practical Guidance states that hourly billing must reflect time actually spent, recognizing that generative AI may reduce the time required for certain tasks. You may bill the time you spend crafting prompts and reviewing outputs. Rule 1.5's prohibition on unconscionable fees already binds, and the AI amendments do not touch Rule 1.5.

Do California courts require disclosure of AI use in filings?

There is no statewide requirement for lawyers, but individual judges impose their own through standing orders, which vary by courtroom. The State Bar's Practical Guidance tells lawyers to check for rules or orders in the relevant jurisdiction that may require disclosure and to comply with them. Read your judge's standing order before every filing.