AB 2575 Was Vetoed. Here's What California's New AI Laws Do Require of Your SNF

TL;DR: Governor Newsom vetoed AB 2575 on September 30, 2026. It is not law. The same day he signed two related laws, AB 1979 and SB 503, and both reach California skilled nursing facilities on January 1, 2027. Next re-check: early January 2027, when both laws take effect, or sooner if the Medical Board or CDPH issues guidance on AB 1979.
AB 2575 is dead. But if you run a California SNF, don't file this under "nothing to do." Two laws signed the same day, AB 1979 and SB 503, apply to health facilities, and SNFs are health facilities. From January 1, 2027, your building has to take reasonable steps to keep licensed nurses in charge of AI-informed clinical decisions, can't let AI direct CNAs in work that requires a license, and has to monitor its AI decision-support tools that carry a bias risk. Your AI vendors, in turn, have to hand you documentation on request. This guide tells you what that means for the Administrator and the DON, and what to do before the date.
What happened to AB 2575, and does any of it still apply?
The Governor returned AB 2575 without his signature. The bill would have let a worker providing direct patient care override a clinical decision support system's output, and barred employers from retaliating against workers for relying on or overriding one. His objection: the anti-retaliation design would put the Labor Commissioner in the position of judging the standard of care for patients, which that office isn't equipped to do.
Does anything carry over? One piece, in narrower form. AB 1979 (below) protects licensed staff's independent judgment. Nothing in the 2026 laws replaces AB 2575's anti-retaliation rule. If you started preparing for AB 2575, the "What isn't required" section below tells you what you can stop worrying about.
What does AB 1979 require of your SNF starting January 1, 2027?
AB 1979 adds Business & Professions Code §22758.5. It does three things that matter on a nursing unit:
- Licensed judgment stays in charge. Your facility must take reasonable steps so a licensed provider, acting within their scope of practice, keeps the ability to exercise independent professional judgment whenever care is informed by the output of a clinical decision support system (CDSS).
- AI can't direct unlicensed staff in licensed work. Your facility can't use an AI tool to direct, guide, supervise or instruct unlicensed personnel in any clinical function the law reserves for a licensed professional. (Supervised trainees are carved out.)
- AI can't do licensed work on its own. Same section: no AI tool independently performing a clinical function the law reserves for a license.
What counts as a CDSS? The law defines it as an AI system that produces a prediction, classification, recommendation, evaluation or analysis that aids clinical decisions about the timing of care, diagnosis or treatment. Practical inference, not a claim about any product: in a SNF, tools that might fit that definition if they use AI include fall-risk or pressure-injury risk scores, sepsis or resident-deterioration alerts, acuity or staffing-need scores, and care-plan suggestions. Whether a specific tool in your building qualifies is a question for your vendor and your counsel. Your move: start the inventory in the checklist below.
The CNA angle. Practical inference, have counsel confirm: the "licensed provider" protection is defined by reference to professionals licensed or certified under Division 2 of the Business & Professions Code. CNAs are certified under the Health & Safety Code, so that first protection most likely speaks to your RNs, LVNs and physicians, not your CNAs. The second rule, no AI directing unlicensed staff in licensed functions, is the one that touches CNAs. That matters in a SNF because CNAs deliver most of the hands-on care.
The documentation carve-out. The law doesn't apply to automated systems used for documentation and communication that doesn't involve the application of professional judgment, such as automated messages, reminders, or helping patients find information. Practical inference: whether a particular AI charting feature "involves the application of professional judgment" is a fact-specific question the statute doesn't answer and no court or agency has tested. Don't assume a tool is inside the carve-out because it sounds like documentation. Ask the vendor how it works, and write down the answer.
Enforcement. The law doesn't name a penalty or an enforcing agency for facilities. It says a violation by a physician goes to the Medical Board or the Osteopathic Medical Board, lets a licensing board seek an injunction where a violation amounts to practicing without a license, and preserves other existing remedies. Silence on facility penalties is not a reason to wait. Treat it as an open question.
What does SB 503 require of your SNF and your AI vendors?
SB 503 adds Business & Professions Code §22758. A SNF that uses a CDSS is a deployer under the law. Your obligations:
- Make reasonable efforts to identify any CDSS you use that has a known or reasonably foreseeable risk of biased impacts: an adverse impact, including diminished access to care, quality of care or outcomes, on someone because of a protected characteristic.
- Regularly monitor those systems, and take reasonable and proportionate steps to mitigate the risk.
Your vendors (the law calls them developers) have their own job. They must make reasonable efforts to mitigate bias risk and give you a statement of intended uses and known risks, plus documentation covering training data (including how representative it is), how the system was evaluated for performance and bias, data governance, intended benefits, known bias risks and mitigations, and recommended use and monitoring. They owe you that on request or at initial sale, whichever comes first, and again after material updates.
Your move: add the documentation request to your next vendor renewal. The law names no penalty or enforcing agency for SNFs either. It does say that complying with it is not a defense to a claim of unlawful discrimination, and that it sits on top of, not in place of, other state AI law.
Does this apply to your facility?
SNFs: yes, on the statutory text. Both laws define "health facility" by pointing to Health & Safety Code §1250, and §1250 includes skilled nursing facilities. Both take effect January 1, 2027: they passed by majority vote with no urgency clause, so the usual rule in Article IV, Section 8 of the California Constitution applies. That's a textual reading. Courts haven't tested it, so treat it as your starting point for a conversation with counsel. Your move: put "AB 1979 and SB 503" on your next call with them.
Assisted living and RCFEs: probably not. Practical inference: RCFEs are licensed outside §1250, so neither law's "health facility" definition reaches them. That's our reading of the text, not a settled answer. If you run an RCFE, confirm with counsel instead of assuming either way.
What should your Administrator and DON do before January 1, 2027?
Practical inference throughout this section: neither law prescribes a checklist. These are sensible steps toward the "reasonable steps" and "reasonable efforts" standards, split by who owns them. Not legal advice.
Administrator (contracts, policy, budget, exposure)
- Build the tool inventory with your DON. List every tool that scores, flags, classifies or recommends, including modules inside your EHR. For each, ask the vendor in writing whether AI is involved.
- Add SB 503 asks to vendor contracts and renewals. Request the vendor's risk statement and documentation now. Make delivery on request and after material updates a contract term.
- Name an owner and a cadence for bias monitoring. Decide who reviews each identified tool, how often, and where the record lives.
- Adopt a short "reasonable steps" policy. Licensed clinicians make the call; tool output is input, not instruction; licensed staff can depart from it.
- Brief your owner or board and your counsel. Bring the inventory and the two statutes.
Director of Nursing (workflow, licensed vs. unlicensed roles, training)
- Walk the floor workflow for AI that tells CNAs what to do. Look at task lists, alerts and prompts. With counsel, decide which tasks in your building are licensed clinical functions. If a tool is directing a CNA in one, that's the AB 1979 problem.
- Keep a licensed review step. Make sure an RN or LVN reviews AI output before it changes a care plan, assessment or order, and that the review is documented.
- Train licensed staff. They can and should depart from a tool's output when their judgment says to, and note why.
- Train CNAs and charge staff. A tool's prompt is not an order. When it seems to be asking for something that needs a license, escalate to the licensed nurse.
- Support the bias review clinically.Tell the Administrator when a tool's output looks different for particular groups of residents, such as by primary language, race or disability.
What isn't required (anymore)?
If you acted on the earlier version of this post, here's what you can stop worrying about:
- No anti-retaliation rule for AI overrides in the new laws. AB 2575's rule died with the veto, and neither AB 1979 nor SB 503 replaces it. Your existing employment obligations are unchanged. This post isn't employment-law advice.
- No required notice to staff. Neither new law requires you to give staff advance written notice of AI tools. SB 503's disclosure runs the other way, from vendor to facility. The 90-day notice you may have been planning for isn't a legal requirement.
- No new vendor-liability rule. Neither law contains a rule about who is liable when a clinician doesn't override an AI output.
Your tool inventory and your vendor conversations are still worth doing. They now serve AB 1979 and SB 503 instead.
Frequently Asked Questions
Is AB 2575 law?
No. Governor Newsom vetoed it on September 30, 2026.
What replaced AB 2575?
Nothing replaced it directly. Two related laws signed the same day, AB 1979 and SB 503, cover different ground: keeping licensed judgment in charge of AI-informed care and barring AI from directing unlicensed staff in licensed work (AB 1979), and monitoring AI decision-support tools for bias and getting documentation from vendors (SB 503). Neither has AB 2575's anti-retaliation rule.
When do AB 1979 and SB 503 take effect?
January 1, 2027. Both passed by majority vote without an urgency clause, so the standard constitutional rule applies.
Do they apply to skilled nursing facilities? To assisted living?
On the statutory text, yes for SNFs: both laws define "health facility" by reference to Health & Safety Code §1250, which includes skilled nursing facilities. Courts haven't tested that reading. Assisted living and RCFEs are most likely outside both laws because they're licensed outside §1250, but that's a practical inference, so confirm with counsel.
Do we have to let nurses override AI recommendations?
AB 1979 requires your facility to take reasonable steps so a licensed provider keeps the ability to exercise independent professional judgment when care is informed by a decision support system's output. It doesn't use the word "override," set a procedure, or create an anti-retaliation protection. In practice, a written policy that licensed staff can depart from a tool's output, with their reasons documented, may help show you've taken those steps. That's a practical inference, not a statutory checklist or a safe harbor.
Can AI tools direct our CNAs?
Not in work the law reserves for a licensed professional. AB 1979 bars a health facility from using an AI tool to direct, guide, supervise or instruct unlicensed personnel in a clinical function that requires a professional license. Which tasks in your building count as licensed functions is a question to settle with your counsel.
What must our AI vendor give us under SB 503?
A statement of intended uses and known or reasonably foreseeable risks, plus documentation on training data, how the system was evaluated for performance and bias, data governance, intended benefits, known bias risks and mitigations, and recommended use and monitoring. It's due on request or at initial sale, whichever is earlier, and after material updates.
Does AB 1979 apply to AI charting tools?
The law doesn't apply to automated systems used for documentation and communication that doesn't involve the application of professional judgment. Whether a given AI charting feature fits that description depends on what the feature actually does, and nobody has tested it. Ask your vendor to explain, in writing, how each feature works.
Disclaimer: This post is informational, not legal advice. AB 2575 was vetoed and is not law. AB 1979 and SB 503 take effect January 1, 2027, and their application to specific tools and workflows is untested. Confirm any compliance planning with your own counsel or state association (for example, the California Association of Health Facilities) before acting on it.
Where Relic Care fits in
Charting Assistant drafts nurses' daily notes from shift conversations, reviewed and signed off by your team before they're final. And if keeping up with laws like these, alongside every other federal, state and facility requirement, is more than one person can track, Compliance Assistant is built to stay current as rules change and to answer with sources.
Whether any specific AI charting feature falls inside AB 1979's documentation carve-out is fact-specific and untested. Using any AI tool doesn't make a facility compliant — ask your vendor to answer that question for your workflow in writing.
Sources
- Office of the Governor, State of California, Veto message, Assembly Bill 2575, September 30, 2026
- Office of the Governor, State of California, Governor Newsom issues legislative update 9.30.2026, September 30, 2026
- Office of the Governor, State of California, California's nation-leading AI framework just got stronger: Governor Newsom signs more first-in-the-nation worker protections and more, September 30, 2026
- California Legislative Information, AB 1979 chaptered text (Ch. 854, Statutes of 2026)
- California Legislative Information, SB 503 chaptered text (Ch. 857, Statutes of 2026)
- California Legislative Information, AB 2575 Official Bill History
- California Legislative Information, Health & Safety Code Section 1250
- California Legislative Information, California Constitution, Article IV, Section 8
More for the People Running Your Facility

Who's Legally Accountable for an AI-Drafted Nursing Note?
CMS's single-signer model and California's author-of-record rule put accountability on the signing clinician — not your AI scribe vendor.

AI in Healthcare Compliance: What It Actually Does for SNFs Right Now
What "AI in healthcare compliance" actually covers today, and the one line California and federal regulators keep drawing: AI can inform, never decide.

Using AI Scribes - How about Patient Consent?
Planning to roll out an AI scribe at your facility? See what CMIA, CIPA, and the federal Wiretap Act require for patient consent before you go live.


