California Now Makes Lawyers Verify Every AI Citation
Matthew Leo · Published October 2, 2026 · Business & Policy
California lawyers can use generative AI, but they cannot hand legal work over to it and trust the result. Governor Gavin Newsom approved Senate Bill 574 on September 30, turning several familiar warnings about AI-assisted legal work into explicit duties.
The law requires an attorney using generative AI to take reasonable steps to verify its output, correct errors and personally check every citation filed with a court. It also requires disclosure when AI is used to create documents submitted to a court.
That matters because the rule is aimed at the ordinary failure modes lawyers already face: invented cases, inaccurate summaries and confidential client information placed in a system with unsuitable access controls.
What the law actually requires
The final text of SB 574 says lawyers may not delegate the practice of law to generative AI. When they use it as an assistant, they must:
- Keep confidential, identifying and other nonpublic information out of systems that do not restrict access to authorized people.
- Take reasonable steps to verify the accuracy of the output, including case and statutory citations.
- Correct erroneous or hallucinated material they use.
- Disclose AI use to the court for documents submitted to the court.
The bill also changes the rules for arbitrators. They may not delegate any part of their decision-making to an AI tool. If they rely on AI-generated information outside the record, they must disclose that use to the parties and, where practical, let the parties comment.
The law takes effect January 1, 2027. Reuters described it as the first US state law focused specifically on lawyers’ and arbitrators’ use of generative AI.
This is more than a warning about hallucinations
The citation requirement is the easiest part to understand, but the confidentiality rule may have the larger day-to-day effect. A lawyer cannot assume that a consumer chatbot, a business subscription and a privately configured legal tool handle prompts in the same way.
The relevant questions are practical: who can access submitted information, whether prompts are retained, whether they can be used for training, where the data is processed and what the firm’s agreement actually promises. A product label that says “enterprise” does not answer those questions by itself.
SB 574 also leaves room for judgment. It requires “reasonable steps” to verify output rather than prescribing one technical process. Firms will still need internal rules that match the sensitivity of the work.
The Canadian position is similar, but not identical
Ontario has not enacted a direct equivalent to SB 574. The Law Society of Ontario’s technology resources instead explain how existing duties apply when lawyers use new tools. Its guidance highlights confidentiality, competence, supervision, candour and accurate billing.
The Canadian Bar Association’s AI guidance is equally clear about verification: lawyers remain responsible for checking information produced by generative AI and following court directives.
The difference is enforcement structure. California has placed specific AI duties in statute, including mandatory disclosure for court documents and an express ban on AI making arbitral decisions. Canadian lawyers currently face a mix of existing professional rules, law-society guidance and court-specific requirements.
For a Canadian firm doing cross-border work, the safest workflow is not to maintain a lower standard at home. Verify every authority from an original legal source, keep protected information out of tools that lack suitable contractual and technical controls, and record enough of the review to show who checked the final work.
Tags: SB 574, legal technology, AI regulation, lawyers, California