Quick answer. California lawyers do not practice under a separate AI ethics code. The State Bar of California’s COPRAC guidance explains that existing duties of competence, confidentiality, supervision, client communication, billing, and candor to the tribunal already govern generative AI use. The practical lesson from the cases is equally straightforward: lawyers may use AI as a tool, but they cannot delegate professional judgment to it, cannot submit unverified citations, and cannot assume that client information entered into a consumer platform will remain privileged or confidential.

This article is written primarily for attorneys. It is intended as a practical risk-management and litigation-integrity guide for California practitioners who are using, supervising, or confronting generative AI in legal work. It focuses on what the governing guidance actually says, where sanctions risk is emerging, and what policies firms should have in place now.

What COPRAC Actually Requires

The State Bar of California’s Standing Committee on Professional Responsibility and Conduct issued its Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law on November 16, 2023. COPRAC did not create a new rule. It explained how existing professional obligations already apply to generative AI. That framing matters because it keeps the analysis disciplined. The operative question is not whether AI is categorically permitted or categorically forbidden. The question is whether the lawyer’s use of the tool complies with the same duties that already govern the practice of law.

Those duties include competence and diligence under Rules 1.1 and 1.3; confidentiality under Business and Professions Code section 6068, subdivision (e), and Rules 1.6 and 1.8.2; supervision under Rules 5.1 through 5.3; communication with the client under Rule 1.4; reasonable fees under Rule 1.5 and Business and Professions Code sections 6147 and 6148; candor to the tribunal and meritorious advocacy under Rules 3.1 and 3.3; and compliance with other applicable law under section 6068, subdivision (a), and Rules 8.4 and 1.2.1.

COPRAC is careful in tone. It does not treat generative AI as inherently improper. It treats it as a tool that presents unusual risks because it can sound confident while being wrong, because providers’ data practices vary, and because lawyers may be tempted to over-rely on outputs that appear polished. That cautionary framing is the right one for any article directed to California lawyers.

The Real Ethical Pressure Points

A. Confidentiality and privilege are not the same thing.

COPRAC makes the confidentiality point directly: a lawyer must not input confidential client information into a generative AI solution that lacks adequate confidentiality and security protections. The guidance expects lawyers to review the provider’s terms, understand whether the tool uses inputs for training or other purposes, anonymize client information where appropriate, and consult IT or cybersecurity professionals before using tools with confidential material.

That does not mean every use of AI destroys privilege. But it does mean that lawyers should stop speaking loosely about consumer AI platforms as though they were the equivalent of a privileged internal work environment. The safer and more accurate practice is to assume that privilege and confidentiality analysis will turn on the specific tool, the provider’s terms, the security environment, the purpose of the communication, and the role of counsel.

B. Competence requires verification, not vibes.

COPRAC states that a lawyer must understand, to a reasonable degree, how the technology works, its limitations, and the governing terms of use. More importantly, the lawyer must critically review, validate, and correct both the input and the output. The duty of competence requires more than merely catching false citations. It requires the lawyer to exercise independent professional judgment over the substance of the work product.

That is the part many lawyers miss. A chatbot may draft prose quickly. It may suggest a useful structure. It may even surface a lead worth checking. But the lawyer still bears responsibility for legal analysis, the choice of authorities, the treatment of adverse law, and the factual accuracy of anything filed or sent.

C. Supervision now includes AI supervision.

COPRAC’s supervision discussion is especially important for firms. Managing and supervisory lawyers should establish clear policies regarding permissible AI use and should train lawyers and nonlawyers on both the practical and ethical risks of these tools. A subordinate lawyer may not rely on a supervisor’s direction as cover for improper use. The same is true of a nonlawyer employee or vendor who uses AI in a way that creates confidentiality, billing, or candor problems.

D. Client communication may be required even without a statewide disclosure rule.

California does not currently have a blanket statewide rule requiring attorneys to disclose every use of generative AI to either clients or courts. But COPRAC expressly advises lawyers to evaluate whether disclosure to the client is appropriate based on the facts and circumstances, including the novelty of the technology, the risks involved, the scope of the representation, and the sophistication of the client. Lawyers should also review any client instructions or outside-counsel guidelines that restrict AI use.

That means AI disclosure is not only a standing-order issue. It can also be a Rule 1.4 issue, an engagement-letter issue, and a client-guidelines issue.

E. Billing rules still apply.

COPRAC also gives practical guidance on fees. A lawyer may bill for the actual time spent crafting prompts, reviewing outputs, editing drafts, and performing the legal work necessary to verify and improve the result. A lawyer may not bill hourly for time saved by the AI. If AI-related costs are passed through, the fee agreement should explain the basis for those fees and costs in compliance with California law.

Heppner and the Confidentiality Warning

United States v. Heppner (S.D.N.Y. Feb. 17, 2026, No. 1:25-cr-00503) has attracted attention because it addresses AI use through the lens of privilege and work product rather than hallucinated citations. The case should be described carefully. It is not a blanket holding that all AI-assisted legal communications are unprivileged. It is better understood as a fact-specific warning that client communications with a consumer AI platform may fall outside privilege or work-product protection depending on the circumstances and the provider’s terms.

For California practitioners, the practical lesson is the same lesson COPRAC already emphasized: do not assume that a consumer chatbot is a secure channel for case strategy, client narratives, or identifying facts. Lawyers should give clients clear written instructions not to use consumer AI tools to analyze their cases unless counsel has affirmatively vetted and approved the platform.

The Hallucination Cases: The Doctrine Is Simple Even When the Docket Is Not

The sanctions cases still matter because they show how courts react when lawyers fail to verify what they file. The precise sanctions, procedural posture, and language vary from case to case, but the common rule is stable: lawyers must read what they sign, verify the authorities they cite, and correct errors immediately once discovered.

Mata v. Avianca, Inc. (2023) 678 F.Supp.3d 443 remains the foundational federal hallucination decision. California lawyers, however, should pay particular attention to Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426, because it is California’s first published appellate decision addressing sanctions for AI-generated fake citations in a filed brief. That makes Noland the most useful California bridge between COPRAC’s guidance and actual consequences in California practice.

The broader sanctions docket also shows that courts often react more harshly to concealment than to the initial error. Negligent failure to verify is bad enough. Doubling down, filing without checking after a warning, or quietly replacing defective work without explaining the problem is what turns a bad mistake into a credibility and sanctions disaster.

California-Specific Developments

California’s Judicial Council adopted Rule 10.430 and Standard 10.80 effective September 1, 2025. Those provisions govern courts, court staff, and judicial officers rather than private attorneys directly. For that reason, lawyers should not overstate them as attorney-conduct rules. But they still matter because they show what California’s courts themselves regard as baseline concerns: confidentiality of nonpublic information, accuracy, hallucination control, bias control, and in some circumstances disclosure of AI-generated content.

That development supports a broader practical point. Even where no statewide attorney disclosure rule exists, lawyers should expect judges to scrutinize AI use through standing orders, local rules, courtroom practice, and ordinary expectations of competence and candor.

Minimum AI Compliance Protocol for California Lawyers

• Verify every case citation, quotation, and record citation against a primary or trusted database before filing or serving the document.
• Do not input confidential or client-identifying information into a public or consumer AI tool unless the platform has been vetted for security, confidentiality, and data-use limitations.
• Adopt written internal policies governing who may use AI, for what tasks, with what approvals, and under what review requirements.
• Train lawyers, paralegals, assistants, and vendors on confidentiality, hallucination risk, supervision rules, and billing limits.
• Consider whether client disclosure is appropriate under the circumstances and review all client guidelines for AI restrictions.
• Document the human review process, especially in high-risk filings, expert work, and motion practice.
• Bill only for actual lawyer and staff time spent; do not bill for time saved by the machine.
• Check the assigned judge’s rules and any applicable local requirements before filing work that involved AI assistance.

What This Means for Law Firms Right Now

For most firms, the question is no longer whether AI will be used. It is whether it will be used in a way that is documented, supervised, and professionally defensible. Firms that treat AI as a forbidden black box will fall behind. Firms that treat it as a substitute for legal judgment will create sanctions and malpractice risk. The sustainable position is in the middle: use it where it adds efficiency, but build human verification, confidentiality controls, and supervisory review into the workflow.

A good firm policy should answer at least five questions. What tools are approved? What information may never be entered into them? What level of attorney review is required before work product leaves the firm? When must the client be informed? And who is responsible for training and enforcement?

Conclusion

The best way to think about generative AI in California practice is not as a novelty and not as a separate ethics regime. It is a technology that intensifies familiar duties. COPRAC says as much. The cases say the same thing in harsher language. Lawyers may use AI, but they must remain competent, protect confidentiality, supervise others, communicate appropriately with clients, bill honestly, and tell courts the truth.

Used carefully, AI can make lawyers faster. Used carelessly, it can make them sanctionable. The dividing line is not whether a machine touched the draft. The dividing line is whether the lawyer exercised judgment worthy of signing it.

Disclaimer. This article is provided for general informational purposes only and is not legal advice. Reading it does not create an attorney-client relationship. Lawyers should analyze the facts of each matter, the governing rules in each jurisdiction, any client instructions, and any judge-specific requirements before relying on AI-assisted work product.