California's New AI Law Will Land on Your Support Team's Desk. Here's What to Sort Out Before January.
If anyone on your team uses an AI tool to summarize medical records, draft a letter or check a citation, a new California law will shape how that work gets done.
SB 574 was signed by Governor Gavin Newsom on September 30, 2026, and takes effect January 1, 2027.
The law’s duties for legal work fall on attorneys. In a PI firm, though, much of the work AI might touch is prepared before an attorney sees it: records summaries, treatment chronologies, demand drafts, letters to providers. That preparation happens on the desks of paralegals, case managers and legal assistants, in the office or remotely. So January is a good deadline for the whole team.
What SB 574 covers
Three parts of SB 574 touch the work support teams prepare every day.
Checking AI output. Attorneys must take reasonable steps to check the accuracy of AI output and correct errors or made up content in material they use. Citations in a court filing must be ones the attorney has "personally verified," whether or not AI was involved in finding them.
Disclosure to the court. Attorneys must disclose their use of generative AI to the court "for all documents submitted to the court." The statute does not spell out what form that disclosure takes.
Confidential information. Attorneys may not enter confidential, personal identifying or other nonpublic information into a generative AI system unless access is limited to the attorney and authorized people bound by confidentiality.
Why PI firms should pay attention
A PI file is full of the kind of information the confidentiality part covers: medical records and bills, treatment histories, diagnoses, addresses, phone numbers, dates of birth, wage and income documents.
It is also full of the kind of work people are tempted to hand to an AI tool: summarizing long records, building treatment timelines, drafting first versions of letters and demands.
That overlap is where things can go wrong. For example, if someone pastes a treatment summary into a public AI chat tool to speed up a demand draft, that client's medical information is now in a system the firm does not control. Nobody has to intend any harm. It only takes a busy afternoon and a tool that seemed helpful.
The same goes for accuracy. If an AI summary misreads a date of service or leaves out a provider, and nobody checks it against the records, the mistake can travel straight into the demand.
Five things to sort out before January 1
These are practical suggestions for getting organized. Your counsel is the right person to tell you what the law requires of your firm.
1. List the AI tools your team uses. Ask everyone, not only the people you expect. Include free chat tools people picked up on their own, browser add ons, and AI features built into software you already pay for, such as document, email or case management tools. A list you can see is a list you can make decisions about.
2. Know where client data goes. For each tool, find out where uploaded information is stored, who can access it, and whether the account belongs to the firm or to one person. If the answer is unclear, keep client information out of that tool until it is clear.
3. Check AI assisted work against the source before it reaches an attorney. A records summary gets compared with the records. A date in a chronology gets matched to the page it came from. A citation gets pulled and read. Make this a named step in the workflow, so it still happens on a busy week.
4. Note AI use on anything headed to court. A short note in the file showing which tool was used on which document, and by whom, makes the attorney's disclosure easier to prepare. It also helps if questions come up later.
5. Ask outside support providers the same questions. If any part of your casework is handled outside the firm, ask how they handle your files, which tools they use, and who on their side can see client information. Clear, specific answers are a good sign. Vague ones are worth following up on.
Where support teams fit
SB 574 makes the attorney responsible for checking AI work. That check goes faster when the work arrives ready for it: a records summary that points to the page each fact came from, a citation already pulled, a note on which tool was used. That preparation is where a supervised support team, in house or outsourced, earns its place.
It also protects what comes next. A demand built on a checked summary is less likely to need fixing later, and a file that does not have to go back for corrections keeps moving toward settlement.
Sources: California SB 574, chaptered text
Ardent is a legal process outsourcing (LPO) provider in the Philippines that focuses only on legal support for law firms and legal professionals. Ardent Paralegal and Business Solutions has supported U.S. personal injury law firms with dedicated legal support teams since 2015. If you've had a difficult experience with outsourcing before, we're happy to walk through what we do differently. No pressure, just a conversation.
When does SB 574 take effect?
January 1, 2027.
Who do its duties apply to?
Attorneys. Separate parts of the law also cover arbitrators.
What does it say about confidential information?
Attorneys may not enter confidential, personal identifying or other nonpublic information into a generative AI system unless access is limited to the attorney and authorized people bound by confidentiality.
Why does it matter to support teams?
Support teams, in house or outsourced, prepare much of the work the law touches. The five suggestions above are a practical place to start.
Comments