What California’s New Law Means for Legal Teams Using AI

California just became the first state to regulate how attorneys use AI by statute, and the new law takes effect January 1, 2027. Here’s what’s in Senate Bill 574, how it affects companies and firms inside California and out, and whether other states could be next.

What’s in CA SB 574

Senator Tom Umberg, chair of the state’s Senate Judiciary Committee, introduced Senate Bill 574 in February 2025. After two rounds of amendments in the Senate and four in the Assembly, it passed both chambers on the last day of the session. On September 30, 2026, Governor Gavin Newsom signed it.

The State Bar first issued AI guidance for attorneys in 2023, and the Judicial Council followed in 2025 with rules for how judges and court staff use generative AI. In May 2026, the Bar updated its guidance to address agentic AI, which can carry out tasks with little human direction.

Now some of those expectations are written into state law. The core theme of the new law is that attorneys can use generative AI “to assist in the practice of law,” but they can’t “delegate the practice of law” to it. SB 574 also sets limits for arbitrators and tightens citation rules for filings in California courts, whether or not AI was involved.

A lot of the law puts in writing what careful lawyers were already doing, like checking citations and protecting client info. The biggest new requirement is telling the court when AI was used. 

Here’s what the law covers:

• Keep sensitive information protected. Attorneys can’t put confidential, personal identifying, or other nonpublic information into an AI system unless access is limited to them and the people they authorize, who also have to keep it confidential. The law’s definition of personal identifying information covers things like dates of birth, witness addresses and phone numbers, and medical and financial details.
• Double-check AI output. Attorneys have to take reasonable steps to verify anything AI produces and fix errors or hallucinations before using it. That goes for any material used by the attorney, contract review included.
• Tell the court when AI was used. Attorneys who use AI have to disclose it on documents they submit to a court. For content shared with the public, they only have to consider disclosing it.
• Check every citation personally. Every citation in a court filing has to be personally verified by an attorney responsible for it, whether it came from AI, a research database, or a colleague’s old brief. 
• Keep arbitration decisions human. Arbitrators can’t hand off any part of their decision-making to AI. If they want to rely on AI-generated information from outside the record, they have to tell the parties first and, where practical, let them weigh in.
• Citation errors can lead to sanctions. Courts can sanction attorneys and their firms, and firms are generally on the hook for their lawyers’ and staff’s violations. Attorneys usually get 21 days to withdraw or fix a challenged filing first. The bill doesn’t spell out separate penalties for the other duties.

Some of the details are still fuzzy, though. The law doesn’t draw a firm line between assisting and delegating or even what type of AI tools need a disclosure. For example, it doesn’t say whether the AI-powered grammar suggestions now built into Word trigger disclosure, or what a disclosure should say. It also doesn’t say how hands-on “personally verified” has to be. An earlier version required attorneys to read every cited source personally, but lawmakers cut that before the bill passed.

Implications for California companies

SB 574 takes effect January 1, 2027, which means legal teams have a little time to plan and adjust their workflows. 

The law applies to every attorney licensed in California, including in-house counsel. The citation rule also reaches anyone filing in a California court, wherever they’re licensed.

Law firms

Law firms will confront SB 574 most in court filings. Firms are generally on the hook for violations by their lawyers and staff, so citation checks and disclosure become firm-wide processes. On large filings with many contributors, the law doesn’t say which attorney is responsible for the citations, so firms will have to decide for themselves.

In-house teams

In-house teams will feel the law in day-to-day legal work. The duties to protect sensitive information and check AI output apply beyond the courtroom, including when in-house lawyers use AI to review contracts or answer questions from the business. A lot of that contract work happens on legal ops and contracting teams. The law only covers attorneys, but an attorney who uses that work is still on the hook for any AI output in it. And since the law doesn’t say how these duties are enforced outside court, problems will likely surface only after something goes wrong.

In-house teams also hire firms that file in court, and a sanctioned filing still lands on their case. Now is a good time to ask outside counsel how they’re checking citations and handling AI disclosure, and to add those expectations to your outside counsel guidelines.

Will other states follow?

California is the first state to write AI rules for lawyers into law, and others could follow. When California passed its consumer privacy law in 2018, it stood alone. Today, roughly 20 states have their own. Lawyer regulation may spread differently, though. In most states, the state supreme court regulates the practice of law, so the next round of AI rules is more likely to come from courts than legislatures. That’s already happening:

• New York: The state court system adopted a statewide rule on attorney AI use in June 2026. Lawmakers have also introduced bills that would require an AI-disclosure affidavit with court filings, though similar bills have stalled in committee.
• Florida: The Florida Supreme Court updated its practice rules so that signing a filing certifies its cited authorities exist and are cited accurately.
• Illinois: The Illinois Supreme Court’s AI policy lets lawyers use AI without disclosing it, as long as they review the output. The court decided its existing rules were enough.

Citation verification is the most likely rule to spread, since California and Florida both landed on it. Disclosure is where states will probably keep disagreeing. For teams working across state lines, that means a patchwork for a while, and some may find it simpler to treat California’s rules as their baseline everywhere.

What SB 574 could mean for the practice of law

SB 574 makes it clear that a lawyer has to answer for whatever AI produces. That idea will likely shape workflows and accountability well beyond California:

Knowing your tools is part of the job. California’s State Bar guidance says understanding an AI tool’s capabilities and limits is part of a lawyer’s duty of competence. It’s also proposed adding AI guidance to several professional conduct rules. AI training is likely to become an expectation for lawyers and support staff alike.

AI agents will blur the delegation line. The difference between assistance and delegation is harder to pin down when agentic tools take on multistep work. The Bar’s 2026 guidance already says the more autonomy a tool has, the more oversight the lawyer owes.

Disclosure could turn into boilerplate. Since the law doesn’t define what counts as using AI, firms might decide to play it safe and add one blanket disclosure line to every filing. If every filing says the same thing, though, the disclosure stops telling courts much. Courts may define what a useful disclosure looks like, or shift the focus to verification. Florida’s Supreme Court already did that, replacing a patchwork of local AI disclosure orders with one rule about whether citations are real.

How to prepare

If you’re an attorney or legal professional in California, the countdown is on to prepare for SB 574 in the new year. Even teams outside the state should start to prepare now, so that any future regulations aren’t a scramble. 

• Audit current AI use in your tools. List every AI tool your team uses, including AI features built into research, review, and contracting software. For each one, find out who can access what you enter, including the vendor. Most vendors publish these details. Here’s how Ironclad handles AI data, for example.
• Revisit your guidelines for 2027. Name who verifies citations on every filing, settle your disclosure language, and spell out which data can go into which tools. In-house teams should share these expectations with outside counsel.
• Pinpoint where AI-driven risk accumulates. Risk builds at handoffs, where everyone assumes someone else checked. It also builds where AI output gets reused. A flawed clause in an AI-drafted template ends up in every contract built from it.

If your team hasn’t decided who’s responsible when AI gets something wrong, start with an AI accountability policy.

---

Ironclad is not a law firm, and this post does not constitute or contain legal advice. To evaluate the accuracy, sufficiency, or reliability of the ideas and guidance reflected here, or the applicability of these materials to your business, you should consult with a licensed attorney.