California Becomes First US State to Legislate Limits on AI Use by Arbitrators and Counsel

On September 30, 2026, California Governor Gavin Newsom signed into law the first US state statute aimed specifically at lawyers’ and arbitrators’ use of generative artificial intelligence (AI), Senate Bill 574. The law takes effect on January 1, 2027.

What the Law Requires of Arbitrators

Senate Bill 574 adds a new Section 1282.1 to the California Arbitration Act, which imposes two obligations on arbitrators:

• No delegation: An arbitrator “shall not delegate any part of their decision-making process to any generative artificial intelligence tool.”
• No undisclosed reliance on material outside the record: An arbitrator may not rely on AI-generated information outside the record without first making disclosures to the parties and, as far as practical, giving them a chance to comment on its use.

This new section of the California Arbitration Act does not displace arbitrators’ existing duties of neutrality, competence and diligence. It defines generative AI broadly, covering systems that produce synthetic text, images, video and audio.

What the Law Requires of Counsel

Senate Bill 574 also regulates attorneys’ use of generative AI:

• No delegation: Attorneys may not delegate the practice of law to generative AI.
• Confidentiality: Attorneys may not enter confidential, personal identifying or other nonpublic information into an AI system unless access is limited to the attorney and authorized persons bound to protect it. This is especially important in arbitration, where confidentiality obligations may arise from the parties’ agreement, applicable institutional rules, or a procedural order.
• Verification: Attorneys must take reasonable steps to verify AI output, including every case and statutory citation, and to correct any erroneous or hallucinated content.
• Disclosure and court filings: Attorneys must disclose AI use in documents submitted to a court. Any paper filed in court must contain only citations the responsible attorney has personally verified.

Many of these requirements codify obligations that flow from attorneys’ existing ethical duties. The disclosure and filing provisions refer to courts, and the statute leaves open whether they also govern submissions to arbitral tribunals, as well as what it means to “delegate” the practice of law.

Diverging Global Trends

Senate Bill 574 codifies many of the same principles discussed in our May 2026 Legal Update on ARIHQ v. Santé Québec. In that case, the Quebec Superior Court annulled an arbitral award for improper delegation of the arbitrator’s decision-making authority to AI. California’s new statute expressly bans such delegation.

The disclosure requirement in Section 1282.1(b)(2) also addresses the due-process concern we identified under English and French law: the risk of an award resting on AI-generated material that the parties never had a chance to address. The counsel provisions likewise reflect the nondelegable duty of verification that courts in the United States, England, and France increasingly enforce.

The author of the California bill put the legislative aim this way: decisions entrusted to arbitrators should be “made by people, not algorithms.” Together, these developments suggest a global shift from guidance toward binding rules, both judicial and statutory.

But the direction of travel is not uniform. As legislatures and courts draw firmer lines around the role of AI in dispute resolution, some arbitral institutions are moving AI closer to the core of decision-making itself. As discussed in our November 2025 Legal Update, the AAA-ICDR has introduced an AI Arbitrator, the first service deployed by a major arbitral institution to generate awards rather than merely assist the tribunal. The AAA-ICDR has confined the tool to two-party, documents-only construction disputes where both parties have consented to its use. Cases that turn on questions of witness testimony or credibility are not eligible. The AI Arbitrator reviews the parties’ submissions and prepares summaries, invites the parties’ feedback on those summaries, and then issues a draft award. A human must then review, validate, and, if needed, revise the AI-drafted award before it becomes final. In our earlier November 2025 Legal Update, we noted that these safeguards (party consent, a narrowly-defined case type, and mandatory human review) could give users the confidence to adopt the service. The open question is whether those same safeguards will satisfy legislatures and courts that now insist the decision remain a human one.

Takeaways

For proceedings seated in California: Arbitrators and counsel should review their AI practices before January 1, 2027. This includes documenting how AI services are used, and confirming that confidential case material exists only in access-restricted, closed systems that preserve confidentiality. An award that shows signs of undisclosed AI involvement or improper delegation of decision-making authority may now face a challenge grounded in statute, as well as procedural fairness. Parties should consider agreeing on AI protocols at the start of each dispute, and should raise AI expectations with the arbitral tribunal early.

For proceedings seated elsewhere: Counsel who are members of the California bar will be expected to abide by their existing ethical duties and those set forth in Senate Bill 574. The new law may also influence institutional guidelines and the expectations of tribunals seated in other jurisdictions.