What Attorneys Using Generative AI Should Do Now ?
7/26/20266 min read
California SB 574: What Attorneys Using Generative AI Should Do Now
Where SB 574 stands (as of August 11, 2026): SB 574 passed the California Senate 39–0 in January 2026 and has cleared its Assembly policy committee. As of August 5, 2026 it sits on the Assembly Appropriations Committee suspense file. The Assembly has until roughly August 31, 2026 to send it to the Governor. It has not been enacted, and its text may still be amended. Treat everything below as guidance on a pending bill, not a statement of current California law.
If you use generative AI anywhere in your practice — research, drafting, summarizing a deposition, cleaning up a brief — SB 574 is the bill to understand this month. Not because it's law yet (it isn't), but because it takes the AI-use duties California attorneys already carry and proposes to write them into statute, with real teeth around court filings.
Here's the useful part: you don't need to wait for the vote. Almost everything SB 574 would require already flows from existing Rules of Professional Conduct and the Business and Professions Code. So the sensible move isn't to guess whether the bill passes — it's to align your workflow with these duties now, and be ready either way. This article walks through what SB 574 would do, where it stands, and gives you a checklist you can put to work today.
Nothing here is legal advice or a substitute for your own professional judgment on a specific matter.
What SB 574 would do, in plain terms
SB 574, introduced by Senator Tom Umberg (chair of the Senate Judiciary Committee), is titled "Generative artificial intelligence: attorneys and arbitrators." It would amend and add sections to the Business and Professions Code and the Code of Civil Procedure — most notably adding a new Business and Professions Code § 6068.1 and amending Code of Civil Procedure § 128.7 (the state's analog to Rule 11 sanctions).
Critically, SB 574 does not ban attorneys from using AI. It sets guardrails. The proposed duties fall into a few clear buckets:
Keep confidential information out of public AI tools. An attorney would have a duty to ensure that confidential, personal identifying, or other nonpublic information is not entered into a public generative AI system.
Verify accuracy and fix hallucinations. An attorney would have to take reasonable steps to verify the accuracy of AI-generated material and correct any erroneous or "hallucinated" output before relying on it.
Don't delegate the practice of law to a machine. An attorney could not delegate the practice of law to generative AI. Judgment stays with the lawyer.
Personally read and verify every citation you file. The § 128.7 amendment would prohibit any brief, pleading, motion, or paper filed in court from containing a citation the responsible attorney has not personally read and verified — expressly including citations produced by AI.
Avoid discriminatory outcomes. An attorney would need to avoid AI-assisted conduct that unlawfully discriminates against or has a disparate impact on protected individuals.
The bill also reaches arbitrators (no delegating decision-making to AI; disclosure obligations for AI-generated information outside the record; and full responsibility for the award regardless of AI assistance) and would require judicial officers to disclose reliance on generative AI in researching or drafting their determinations.
One drafting nuance worth flagging
The confidentiality duty is tied to public generative AI systems, and the bill does not spell out a bright-line definition of "public generative AI system." Commentators have noted this leaves a gap for "private" or enterprise-closed tools. Don't over-read that gap: your existing confidentiality duty under Business and Professions Code § 6068(e) and Rule 1.6 doesn't switch off just because a tool is "private," and a client's own policies or engagement terms may restrict AI use regardless.
Why this matters even if the bill stalls
Here's the framing most coverage misses. SB 574 largely codifies duties California attorneys already have. Map it to the rules you already follow:
The California State Bar's COPRAC has already been developing practical guidance and proposed Rules of Professional Conduct changes addressing generative AI, and California Rule of Court 10.430 directs courts to adopt AI-use policies for judges and staff. SB 574 would convert much of that advisory posture into enforceable statute — but the underlying obligations are here now. That's why the practical answer is to build the workflow today.
The motivation isn't abstract. In Noland v. Land of the Free, L.P. (2025) 114 Cal.App.5th 426, the Second District Court of Appeal (Division Three) issued California's first published opinion on AI hallucinations in court filings — sanctioning the appellant's attorney $10,000, referring him to the State Bar, and publishing the opinion expressly to warn other lawyers after finding that 21 of 23 case quotations in his briefs were fabricated by generative AI he had not verified. Here's the part that matters for SB 574: the Noland court's own admonition — that no filing should contain a citation, "whether provided by generative AI or any other source," that the responsible attorney hasn't personally read and verified — reads almost exactly like the bill's proposed § 128.7 amendment. SB 574 would essentially codify what a California appellate court has already told attorneys. Reporting also documents a rise in fabricated-citation filings nationwide, and even seasoned lawyers have been caught. SB 574 is the legislature's response to that pattern.
The SB 574 compliance checklist
Use this to align your practice now. It's organized by where AI actually enters your workflow.
Intake & confidentiality
Identify which AI tools your firm uses and classify each as "public" vs. enterprise/closed, and document the distinction.
Prohibit entry of confidential, personal identifying, or nonpublic information into public AI tools — by policy, not just habit.
Treat as sensitive at minimum: names/contact details of parties, victims, and witnesses; dates of birth; driver's license and Social Security numbers; medical, psychiatric, and financial information.
Confirm client engagement terms and any client-side AI restrictions before using AI on their matter; get informed consent where appropriate.
Research & drafting
Treat AI output as a lead, never as authority. Verify every proposition against a primary source.
Build a "correct-the-hallucination" step into your review: check that cited cases exist, say what the draft claims, and remain good law.
Keep the legal judgment — analysis, strategy, conclusions — with a licensed attorney. Don't let the tool make the call.
Court filings (the high-risk zone)
Personally read and verify every citation in anything you file — including citations you didn't generate with AI.
Retire any "I relied on staff/software" defense; under the proposed § 128.7 change it wouldn't help you.
Pull and confirm each authority in its original source before signing.
Firm policy & supervision
Put an AI-use policy in writing and train everyone who touches client work (Rule 5.1/5.3 supervision).
Log which tools are approved, for what tasks, and with what data limits.
Consider whether AI-use disclosure is appropriate for public-facing content you create.
Review anti-bias/anti-discrimination exposure where AI touches decisions affecting people.
How SB 574 differs from ABA guidance
Attorneys sometimes conflate the two. Keep them distinct:
ABA Formal Opinion 512 addresses lawyers' use of generative AI under the ABA Model Rules. It's influential guidance interpreting model rules — not binding law anywhere by itself.
SB 574 would be a California statute — binding on California attorneys, with the citation-verification piece wired into § 128.7's sanctions framework.
If you practice in California, your governing authority is California's Rules of Professional Conduct, the Business and Professions Code, and (if enacted) SB 574 — informed by, but not replaced by, ABA opinions. Always distinguish jurisdiction-specific authority from national commentary.
What happens next
Watch three things before the end of August: the Assembly Appropriations Committee's mid-August suspense-file hearing (the vote-only session that decides which fiscal bills advance), any further amendments to the bill text (the confidentiality and § 128.7 provisions are the ones to reread), and, if it advances, a floor vote and the Governor's decision. The 2025–26 session ends August 31, 2026.
FAQ
Is SB 574 law in California?
No. As of August 11, 2026 it is pending in the Assembly, on the Appropriations Committee suspense file, after passing the Senate 39–0. It has not been enacted.
Does SB 574 ban attorneys from using AI?
No. It sets guardrails around confidentiality, accuracy, citation verification, non-delegation, and non-discrimination. Attorneys could still use generative AI to assist their work.
What's the biggest practical change for litigators?
The proposed Code of Civil Procedure § 128.7 amendment: you could not file any paper containing a citation you haven't personally read and verified, including AI-generated citations. It removes the "the tool or my staff did it" excuse.
Does the confidentiality rule apply to private or enterprise AI tools?
The bill's confidentiality duty is framed around public generative AI systems and doesn't clearly define that term. But your existing confidentiality duties, and client requirements, can still restrict private-tool use — so don't treat "private" as a free pass.
Do I need to do anything now if I'm not in California?
SB 574 is California-specific, but the duties it codifies track obligations most jurisdictions already impose through their rules of professional conduct. The checklist above is good practice regardless.
What should I do today?
Align your intake, drafting, filing, and supervision workflows with the checklist above. Those steps largely restate duties you already carry, so they hold up whether or not SB 574 passes.