California Regulates Generative Artificial Intelligence in Legal Practice Under SB 574
California Governor Gavin Newsom signed SB 574 on September 30, 2026, establishing the United States‘ first state-level regulations governing attorneys’ use of generative artificial intelligence.
Prohibition Against Delegating Legal Practice to Artificial Intelligence
Senate Bill 574 creates Business and Professions Code section 6068.1, which draws a sharp line between assisting with legal work and delegating core responsibilities. While the statute permits attorneys to use artificial intelligence tools to assist in the practice of law, it bans handing over the actual practice of law to automated systems. The legislation does not explicitly define the exact threshold where assistance crosses into delegation. California attorneys must take reasonable steps to review artificial intelligence outputs, correct inaccuracies or hallucinations, and ensure professional judgment governs all work product, whether or not the documents are submitted to a court.
This statutory requirement operates independently of existing rules governing court documents. Lawyers remain subject to long-standing duties of competence, confidentiality, and professional judgment, which the California State Bar addressed in guidance initially issued in 2023 and updated in May 2026.
Mandatory Personal Verification for Court Citations
Litigation teams face a separate, unconditional mandate regarding citations in court documents. Effective January 1, 2027, briefs, motions, pleadings, and other filings cannot contain citations that the responsible attorney has not personally verified. This obligation stems from California Code of Civil Procedure section 128.7, which already requires a reasonable inquiry into the factual and legal support for court submissions. SB 574 adds a strict personal verification requirement that applies regardless of whether the citation originated from an artificial intelligence tool, a standard legal research database, or another human attorney.
Earlier drafts of the legislation mandated that citations be personally read and verified, but the California Legislature removed the word “read” before enactment. This change eliminates an explicit reading requirement while preserving the underlying obligation to verify. The rule applies to the attorney responsible for submitting the filing, though the text leaves open questions regarding how responsibility is allocated when multiple lawyers contribute to a single brief. Violations of this provision trigger section 128.7’s existing sanctions procedures.
Strict Data Privacy Rules for Witness and Party Information
Internal investigations and civil litigation practices must adapt to strict new data security boundaries. Section 6068.1 prohibits attorneys from entering confidential, personally identifying, or other nonpublic information into generative artificial intelligence systems unless access is strictly restricted to the attorney and authorized persons bound by confidentiality obligations.
Crucially, the statute contains no general exception for personally identifying information obtained from public records. Attorneys frequently use a witness’s date of birth or address to distinguish individuals with similar names, check prior legal proceedings, and investigate credibility. Entering such details into an artificial intelligence tool without proper access restrictions violates the law, even when the underlying data is publicly available.
Court Disclosure Mandates and Practical Compliance Uncertainty
Attorneys who utilize generative artificial intelligence must disclose that use to courts for all submitted documents. However, the text of SB 574 leaves practical mechanics unresolved by failing to specify the required form, scope, or exact triggers for disclosure. Using an automated system to draft a complete legal argument raises different questions than using it to organize research, summarize documents, or edit language. The statute does not clarify whether every supporting activity triggers an identical disclosure obligation.
This ambiguity creates coordination challenges when junior associates use automated tools for research or drafting without the submitting partner’s full knowledge of the technology’s application. Enterprise artificial intelligence subscriptions alone do not establish compliance, as legal departments must confirm user access controls and binding confidentiality terms. Baker McKenzie advises corporate legal departments and outside counsel to establish clear internal protocols before the January 1, 2027 effective date to determine which tools are permissible, who performs citation verification, and how court disclosures are managed.
Frequently Asked Questions
Does SB 574 ban California attorneys from using generative artificial intelligence entirely?
No. The statute expressly permits attorneys to use artificial intelligence to assist in the practice of law, but it strictly prohibits delegating the practice of law to automated systems.
When does SB 574 take effect?
The law becomes effective on January 1, 2027, giving legal teams time to establish compliance protocols and review their software tools.
Does the personal citation verification rule apply only to artificial intelligence outputs?
No. The requirement to personally verify citations in court filings applies to all citations regardless of whether they were generated by artificial intelligence, sourced from legal research databases, or prepared by human colleagues.
Are publicly available records exempt from the data privacy restrictions?
No. The statute prohibits entering personally identifying information—such as a witness’s address or date of birth—into artificial intelligence systems without restricted access and confidentiality safeguards, even when that information originates from public records.
Worth a look