California Bill Would Set AI Guardrails for Attorneys and Arbitrators
California's SB 574 would impose specific duties on attorneys and arbitrators who use generative AI, addressing confidentiality, accuracy, bias, and decision-making authority. The bill amends the Business and Professions Code and Code of Civil Procedure but does not ban AI use-it establishes boundaries for how legal professionals can deploy the technology.
What Attorneys Must Do
Protect confidential information. Attorneys cannot input client data into public AI systems. This includes driver's license numbers, Social Security numbers, dates of birth, contact information for parties and court staff, medical records, financial data, account numbers, and anything sealed by court order or marked confidential by law.
Verify accuracy. Attorneys must check all AI-generated content for errors and hallucinations, remove biased or offensive material, and correct inaccurate output in work prepared by others on their behalf.
Avoid discrimination. Attorneys cannot use AI in ways that unlawfully discriminate against or harm people based on age, race, gender identity, disability, national origin, religion, sexual orientation, socioeconomic status, or other protected classifications.
Check every citation. Attorneys must read and verify each citation in court filings, including those generated by AI. The bill treats AI-generated citations the same as attorney-written ones.
Disclose AI use where appropriate. Attorneys should consider whether to tell the public when AI creates content they share externally.
What Arbitrators Must Do
Make decisions themselves. Arbitrators cannot delegate any part of decision-making to AI. They must independently analyze facts, law, and evidence without letting AI systems influence procedural or substantive choices.
Disclose outside information. If arbitrators use AI-generated information not in the record, they must tell the parties and allow comment when practical. They cannot assume an AI tool's sources are accurate or independently verifiable if the tool cannot cite them directly.
Own the award. Arbitrators bear full responsibility for awards regardless of any AI assistance used in drafting or analysis.
What This Means for Legal Work
The bill treats AI as a tool that must remain under attorney and arbitrator control. It does not prevent legal professionals from using generative AI for research, drafting, or analysis. Instead, it requires human judgment at critical points: verifying citations, checking for bias, protecting confidential information, and making final decisions.
For attorneys, the requirements align with existing professional responsibility rules around accuracy and client protection. For arbitrators, the restrictions prevent AI from replacing their role as neutral decision-makers.
As the bill moves through the Assembly, legal professionals should review how their current AI workflows handle confidentiality, citation verification, and bias detection. Those working in litigation or arbitration face the most direct obligations under the proposed rules.
For those wanting to deepen their understanding of AI in legal work, resources on AI for Legal and the AI Learning Path for Paralegals cover document review, research, and compliance areas that SB 574 addresses.
Your membership also unlocks:
