Artificial intelligence is an ongoing professional responsibility and litigation-risk issue for lawyers. The California State Bar recently published new guidance. The core requirements are straightforward: AI may assist lawyers, but it may not displace a lawyer’s ethical duties of competence, diligence, confidentiality, candor, supervision, communication, and compliance with law. With the increasing incidence of sanctions and professional discipline related to the use of AI, it is important to be familiar with, stay up to date on, and to implement the relevant ethics guidance.

The California State Bar’s new guidance on AI replaces its 2023 guidance and addresses both generative AI and agentic AI. Generative AI (GenAI) creates new content in response to prompts based on patterns learned from large volumes of data. One concern is that GenAI outputs may include plausible but inaccurate information, including hallucinations. Agentic AI presents an additional risk because it can autonomously perform tasks or “workflows” for lawyers without continuous prompting, such as preparing pleadings or discovery responses, conducting document review, or handling client intake. Because agentic systems can execute goals and plans without human intervention, they heighten the risk that the exercise of a lawyer’s judgment may be bypassed unless the lawyer maintains meaningful review and control.

Reliability concerns are not theoretical. A March 2025 Stanford University study titled “Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools” examined LexisNexis Lexis+ AI, Thomson Reuters Ask Practical Law AI, Westlaw AI-Assisted Research, and GPT-4, and manually reviewed outputs for accuracy and fidelity to authority. The study reported that AI research tools made by LexisNexis and Thomson Reuters “each hallucinate between 17% and 33% of the time.” Identifying these errors often requires subtle analysis of the underlying source material, since problems can include poor legal reasoning, claims that a case stands for a proposition it does not discuss, or presentation of a rejected argument as a case holding. 

The new California guidance addresses lawyers’ duty of competence, providing that lawyers must maintain technological competence, preserve professional judgment, and retain control over strategic decision-making, meaning that a lawyer cannot delegate professional judgment to AI. The 2026 update emphasizes that a competent and diligent lawyer must still read and reason and provides that no paper filed in court should contain citations that the attorney responsible for the filing has not personally read and verified.

Confidentiality presents a separate, equally important concern. Data entered into AI systems may be shared, AI systems may lack adequate security, and breaches may occur in multiple ways, such as through prompt content or uploaded materials. The update states that a lawyer may not input “confidential information” into AI absent informed client consent regarding the risks. Lawyers must understand the relevant terms of use, privacy policies, and vendor documentation before using AI in matters involving client or confidential information. 

Lawyers also must ensure that AI use complies with applicable law, including tribunal rulings, privacy law, cybersecurity law, cross-border data transfer law, and intellectual property law. This point reinforces that AI adoption is not only a technology issue but also a legal-risk and compliance issue. 

Supervision is another area addressed in the recent guidance. Lawyers should establish clear policies regarding the use of generative AI and agentic AI, and those policies must be updated as AI systems change, become more autonomous, and are integrated into legal workflows. Periodic training should be conducted to ensure compliance with those policies. 

Client communication should also be considered at the outset of representation and as the representation develops. Lawyers should consider whether to disclose their intent to use AI, how AI will be used, and the risks and benefits of that use, and must follow client instructions that restrict or limit the use of AI. In practice, this means firms should consider whether to discuss AI use in engagement terms, matter protocols, outside counsel guidelines, and client-specific technology restrictions. 

Billing practices require equal care. Billing must reflect the actual time spent on a matter. The guidance provides that general subscriptions for AI tools typically constitute overhead. Lawyers may charge for AI costs specifically incurred for a client matter beyond general office operations, but the fee agreement should disclose those costs. Fees must reflect the lawyer’s actual costs without a markup or profit element unless the client gives informed written consent. 

The duty of candor to the court remains nondelegable. Lawyers cannot delegate their duty of candor to AI, and they remain professionally responsible for truthfulness and legal sufficiency. Lawyers must review all outputs for accuracy and independently verify them. Lawyers must employ only means consistent with the truth, in accordance with the trust placed in them and must not mislead a court by artifice or false statement of law. Lawyers may not present a claim or defense unsupported by law, make a false statement of law, or knowingly misquote the language of a book, statute, decision, or other authority.