News analysis · Published
California SB 574: What Legal AI Governance Should Include
By the ELYMENT AI editorial team · Free to read
California lawmakers approved Senate Bill 574 on 31 August 2026, sending it towards Governor Gavin Newsom for signature or veto. If enacted, the measure would set explicit duties for lawyers using generative AI, including human verification, disclosure and protection of confidential information, while restricting arbitrators from delegating decisions. The bill is not yet law. Its practical signal is broader: professional firms should treat AI as supervised assistance and keep a qualified person accountable for evidence, judgement and client outcomes.
What California lawmakers approved
The California Legislature's official history says the Senate concurred in Assembly amendments to SB 574 on 31 August 2026 by 39 votes to zero and ordered the measure to engrossing and enrolling. Reuters reported on 1 September that both chambers had approved the current version and that it now goes to Governor Gavin Newsom, who can sign or veto it. Until that process is complete, businesses should describe SB 574 as passed legislation awaiting gubernatorial action, not as an enacted law.
The bill would prohibit an attorney from delegating the practice of law to generative AI. It would require reasonable steps to verify AI-assisted material and correct erroneous or hallucinated output. It also addresses confidential and nonpublic information, discriminatory effects, disclosure of AI use in court submissions and personal verification of citations. For arbitrators, it would prohibit delegating any part of decision-making to an AI tool and require responsibility for the final award.
The important distinction is assistance versus accountability
SB 574 does not propose a general ban on AI-assisted drafting, research or workflow support. The operating boundary is whether a qualified professional remains responsible for the work. A system can prepare, classify or suggest; it should not become the unaccountable decision-maker for legal advice, a filing or an arbitral outcome.
That distinction is useful beyond law. Accountants, advisers, insurers and regulated service firms can apply the same pattern: define which tasks AI may assist, which evidence a reviewer must inspect and which outcomes require a credentialled person to approve. A human name on a workflow is not enough if the reviewer cannot see the source material, model contribution and unresolved uncertainty.
Verification, disclosure and confidentiality need workflow controls
The bill's duties translate into product requirements. Citation verification needs access to the cited authority and a recorded reviewer decision. Disclosure needs a reliable way to identify when generative AI contributed to a court document. Confidentiality needs approved tools, data classification, access controls and technical restrictions that prevent sensitive material entering an unauthorised system.
These controls should operate before submission, not as a policy reminder after an error. The workflow should stop when a source cannot be verified, required disclosure is missing or the selected AI service is not approved for the data. Logs should preserve enough evidence to explain the process without unnecessarily retaining privileged or personal information.
Use a five-part professional AI control test
Professional firms can turn the bill's themes into a practical release gate:
- Task boundary: state what AI may prepare and what judgement it must never make or approve.
- Evidence check: require the reviewer to open and verify every material source, citation and calculation.
- Data boundary: approve systems by information class and block confidential or nonpublic data from unauthorised tools.
- Disclosure rule: define when AI use must be recorded, disclosed to a court or client, or included in the work file.
- Decision owner: identify the qualified person who accepts the outcome and can explain, correct or withdraw it.
What business leaders should do next
Legal and compliance leaders should monitor the governor's action and obtain California-specific advice if the bill is enacted. Operationally, firms do not need to wait. Select one AI-assisted professional workflow, trace the data and sources it uses, then test whether a reviewer can identify AI involvement, reproduce the evidence and stop an unsupported output before it reaches a client or authority.
ELYMENT AI's Gemini Enterprise for Legal analysis explains why governed connectors and traceable evidence matter. Our AI document-intake guide shows how to keep consequential record changes behind human approval, while the agent-approval framework helps teams place review at the right boundary. ELYMENT AI can help turn professional obligations into auditable workflow controls without treating policy as a substitute for system design.
This article is general business analysis and not legal advice. The final text, effective date and obligations will depend on whether SB 574 is enacted and on any later guidance or rules.
Sources
- California Legislative Information: SB 574 bill text (2026-08-21) - Official amended bill text covering proposed duties for attorneys, citation verification, confidentiality, disclosure and restrictions for arbitrators.
- California Legislative Information: SB 574 history (2026-08-31) - Official legislative history recording passage, concurrence in Assembly amendments and the 39-0 Senate vote on 31 August 2026.
- Reuters: California lawmakers pass bill governing lawyers' use of AI (2026-09-01) - Independent reporting on the bill's passage, pending gubernatorial action, proposed requirements and professional context.
Continue learning
Related analysis
- California SB 574 Is Law: Make Legal AI Verification Traceable
California has enacted SB 574 for legal AI use. Learn how to build a traceable record for citations, confidentiality, correction and human review.
Frequently asked questions
Is California SB 574 already law?
No. Both legislative chambers approved the current version on 31 August 2026, but as of 2 September it still awaited Governor Gavin Newsom's signature or veto.
Would SB 574 ban lawyers from using generative AI?
No general ban is proposed. The bill would prohibit delegating the practice of law to AI and would impose duties including verification, confidentiality protection and disclosure.
What is the main business lesson from SB 574?
Use AI as supervised assistance: define task and data boundaries, verify evidence, record required disclosure and keep a qualified person accountable for consequential outcomes.