California’s new law does not prohibit lawyers from using generative AI. It raises the bar for how they use it—and puts human judgment, verification, and accountability firmly at the center of legal practice.
Generative AI is rapidly becoming part of everyday legal work. Attorneys are using it to research legal questions, draft contracts, summarize depositions, review discovery, and prepare litigation documents.
But what happens when an AI-generated legal argument sounds convincing, cites genuine cases, and still gets the law wrong?
Until recently, the answer was largely framed around existing professional ethics obligations and guidance from state bar associations.
California has now taken a significant step further.
On September 30, 2026, Governor Gavin Newsom signed Senate Bill 574 (SB 574), making California the first state to enact legislation specifically regulating how attorneys use generative AI in legal practice. The law takes effect January 1, 2027.
SB 574 does not discourage the adoption of AI. It establishes an important principle: AI can assist in performing legal work, but it cannot assume the professional responsibilities of an attorney.
That distinction has far-reaching implications for how law firms select AI tools, supervise their use, and verify the work they produce.
What SB 574 Actually Requires
The legislation builds on existing duties of competence, diligence, and confidentiality while establishing specific statutory requirements for generative AI.
Four provisions deserve particular attention.
1. Attorneys cannot delegate the practice of law to AI
AI can support legal research, analysis, drafting, and document review. But the attorney must retain responsibility for exercising professional judgment.
The statute does not precisely define the boundary between AI assistance and impermissible delegation. That makes meaningful human oversight especially important.
2. Client confidentiality must be protected
Attorneys cannot enter confidential, personally identifying, or other nonpublic information into an AI system unless access to that information is restricted to the attorney and authorized persons bound by confidentiality obligations.
This is more than a question of whether a product advertises enterprise-grade security. Firms need to understand who can access information entered into their AI systems, including vendors and service providers.
3. AI-generated work must be verified
Attorneys must take reasonable steps to verify the accuracy of AI outputs, including case and statutory citations, and correct erroneous or hallucinated material.
The legislation also amends California Code of Civil Procedure section 128.7 to require an attorney responsible for submitting a covered court filing to personally verify its citations.
Importantly, this personal citation-verification requirement applies whether the citations were generated by AI or obtained through traditional research. Section 128.7 retains exclusions for specified discovery-related materials.
4. AI use must be disclosed to courts
The law requires attorneys using generative AI to disclose its use to the court for documents submitted to the court. Attorneys must also consider disclosure when AI is used to create public-facing content.
The statute does not prescribe a uniform disclosure format, leaving practical questions about implementation that firms will need to address.
SB 574 also prohibits arbitrators from delegating decision-making to AI and restricts reliance on AI-generated information outside the arbitration record without appropriate prior disclosure and an opportunity for parties to comment, as far as practical.
The Real Challenge: Verification Is More Than Checking for Hallucinations
Much of the discussion about AI risks in legal practice has focused on fabricated citations—cases that do not exist but appear convincingly in AI-generated briefs.
SB 574 brings attention to a broader problem.
Consider a hypothetical scenario.
An associate uses generative AI to prepare a litigation brief containing fifteen case citations. The attorney checks the citations and confirms that every case exists.
The brief is filed.
Later, opposing counsel discovers that three of the cited cases do not actually support the legal propositions attributed to them.
No cases were fabricated. Yet the legal analysis was flawed.
This illustrates why verification must go beyond checking whether an authority exists.
Attorneys should evaluate whether a cited authority is relevant, accurately characterized, supports the stated proposition, and applies to the facts of the matter.
A useful approach is to verify AI-assisted legal work through four lenses:
- Authority: Is the source authentic, current, and accurately represented?
- Fit: Does the authority support the legal proposition and apply to the issue?
- Facts: Are the underlying facts accurate and supported by the record?
- Judgment: Has the attorney independently assessed the reasoning and conclusions?
This four-lens framework is a recommended practice, not a statutory checklist.
It also reinforces a critical distinction: AI can accelerate the production of legal work, but a faster draft does not necessarily mean a faster, or less demanding, professional review.
From AI Experimentation to AI Governance
For law firm leaders, the significance of SB 574 extends beyond individual attorneys checking their work.
Imagine a firm where associates use ChatGPT for preliminary research, paralegals use AI to summarize discovery, and partners use specialized legal AI platforms to review contracts.
Who determines which tools are acceptable? Who verifies the output? How does the firm know when AI has been used in preparing a court filing?
These are not simply technology questions. They are questions of professional accountability and operational governance.
The answer is not necessarily to impose an elaborate compliance bureaucracy. Particularly for solo and small law firms, the goal should be a practical set of safeguards that fit the firm’s size, practice areas, and exposure to risk.
The key is to move from informal individual experimentation toward consistent, supervised AI-assisted workflows.
Five Steps Law Firms Should Take Before January 1, 2027
1. Take inventory of AI tools
Identify where generative AI is being used across research, drafting, document review, client communication, and administrative workflows. Include tools used by support staff and outside vendors.
2. Review confidentiality protections
Evaluate vendor terms, access permissions, data retention, and model-training practices. Do not assume that an enterprise subscription automatically satisfies confidentiality requirements.
3. Establish verification protocols
Define how attorneys will review AI-generated legal analysis, factual assertions, and citations. For covered court filings, explicitly account for the attorney’s personal citation-verification obligation.
4. Develop an AI disclosure process
Establish a practical way to identify when generative AI has contributed to court-submitted documents and how the firm will make required disclosures, taking applicable court rules into account.
5. Train the entire legal team
Ensure attorneys, paralegals, and relevant support staff understand acceptable AI use, confidentiality restrictions, verification expectations, and escalation procedures when AI outputs appear unreliable.
These measures are recommended implementation practices rather than five separately mandated statutory procedures. They can help firms operationalize the duties imposed by SB 574 and reduce exposure to sanctions, disciplinary proceedings, and reputational harm.
The Bigger Picture: Responsible AI Adoption Becomes a Professional Capability
There is a tendency to view new AI regulation primarily through the lens of compliance costs and restrictions.
But SB 574 also presents an opportunity.
Law firms that learn how to combine AI-assisted productivity with reliable verification, confidentiality safeguards, and professional oversight may be better positioned to deliver efficient legal services without compromising quality.
That matters particularly for smaller firms seeking to compete with larger practices that have greater resources for legal technology.
The competitive advantage will not necessarily come from adopting the most AI tools. It may come from knowing which tasks to entrust to AI, how to evaluate its outputs, and when human judgment must take precedence.
SB 574 may be the first statute of its kind, but it is unlikely to be the last word on AI accountability in the legal profession.
The larger shift is already underway: as AI becomes embedded in professional workflows, the ability to govern its use becomes as important as the ability to use it.
Final Thought
January 1, 2027, should not simply be another compliance deadline on a law firm’s calendar.
It is an opportunity to examine how legal work is changing—and whether existing practices for supervision, verification, and professional accountability are keeping pace.
The question goes beyond just How much legal work can AI help us perform?
It is also How do we ensure that the work remains accurate, defensible, and under meaningful human control?
AI may change how legal work gets done. SB 574 reinforces who remains accountable for it.
Sources and Further Reading