State Bar AI Opinions: what it asks, and the records to keep

State bars generally let lawyers use generative AI if they meet existing duties: competence, confidentiality, client communication, supervision, candor to courts and reasonable fees. ABA Formal Opinion 512 (July 2024) sets the national baseline, and bars including California, Florida, Texas and the New York City Bar have issued their own guidance. Details differ, especially on client consent, disclosure and billing for AI costs, and California's 2026 guidance also covers agentic AI. Read your own state's opinion, because it controls. Read the full answer

California's 2026 practical guidance, which replaced its 2023 version, covers confidentiality, competence, client communication, candor and fees, and adds duties for agentic AI. AI costs billed to a client must appear in the fee agreement and reflect actual cost, with no markup unless the client gives informed written consent. Florida Opinion 24-1 (2024) addresses confidentiality and client consent, supervision, fees and client-facing chatbots. Texas Opinion 705 (2025) stresses competence, confidentiality and checking output, and bars billing for time the AI saved. All three keep the lawyer responsible. Read the full answer

There is no single national rule. ABA Opinion 512 says lawyers need a client's informed consent before putting client information into a self-learning AI tool that could expose it, and should tell clients about AI use when they ask or when it matters to decisions in the representation; boilerplate in an engagement letter is not enough. California's 2026 guidance says lawyers must consider telling clients they plan to use AI. For courts, check local rules and judges' standing orders, some of which require disclosure or certification of AI use. Read the full answer

Start with the vendor's data terms: whether inputs are stored, who can access them, and whether they train the vendor's models. ABA Opinion 512 requires informed client consent before client information goes into a self-learning tool that could expose it. Then review security, the tool's accuracy and error patterns on your own kinds of matters, how outputs will be checked, and the supervision and billing effects. Ask for test results rather than marketing claims; California's 2026 guidance says generalized marketing assurances are not enough to rely on. Read the full answer