Relic Care
Book a Demo

What California's AI Documentation Bill Would Require

Aina Malkonian . MHA | RCFE profile photo
Aina Malkonian . MHA | RCFELead - Long Term Care
AI in CareCompliance
A physician writing on a clipboard beside an open laptop, overlaid with holographic document and padlock icons symbolizing secure, compliant clinical documentation

Current as of July 28, 2026. AB 2575 — California’s "Health Care by Humans Act" — is still a pending bill. It has not been signed into law, and everything below describes what it would require if enacted, not a current legal obligation. Next re-check date: August 10, 2026, right after the bill’s next expected Senate Appropriations action.

AB 2575 would require California health facilities — including, on a textual reading of its own definitions, skilled nursing facilities — to give clinical staff written notice before deploying an AI or clinical-decision-support tool, guarantee staff the right to override that tool’s output, and bar AI vendors from using an unoverridden output as a complete defense against a harm claim. As of July 28, 2026, none of this is law yet: AB 2575 has passed the Assembly and cleared two Senate policy committees, and now sits in the Senate Appropriations Committee awaiting its next action, expected around early August 2026.

Why AB 2575 matters for your facility right now

AB 2575, introduced by Assemblymember Liz Ortega (D-Hayward) and carrying the short title "Health Care by Humans Act," would add three new provisions to California law: Civil Code Section 1714.48, Health and Safety Code Section 1339.76, and a new Labor Code article beginning at Section 2820. Together they build one idea — a human has to stay in the loop when an AI tool touches patient care — into a disclosure duty, a worker-override right, and a liability rule (AB 2575 bill text, California Legislative Information).

The bill has moved fast this session. It passed the Assembly floor on May 27, 2026 by a vote of 48–15, cleared the Senate Labor, Public Employment and Retirement Committee on June 17 (8–2) and the Senate Privacy, Digital Technology, and Consumer Protection Committee on June 24 (4–1), and was referred to the Senate Appropriations Committee on June 29, 2026, 6–2. It’s co-sponsored by the California Federation of Labor Unions, AFL-CIO and the California Nurses Association, and opposed by a hospital-industry coalition coordinated through the California Hospital Association, which has argued in dated committee letters that the bill "sweeps a broad range of clinical technology into … [an] onerous regulatory framework" (CalHospital coalition opposition letter, Assembly Appropriations Committee, April 29, 2026). That fight over amendments is still active, which is part of why nothing here should be treated as final.

What would AB 2575 require your facility to disclose, and when?

Disclosure checklist.

Checklist graphic titled "What AB 2575 would require you to disclose," listing required disclosures — tool developer, funding source, foundation model, intended use, risks, and override rights — and when notice is required: 90 days before deployment, at hire, and annually. A footer note reads that AB 2575 is a pending bill, not yet law.

An AI scribe is a tool that listens to or reads a clinical encounter and drafts the note automatically; a clinical decision support system (CDSS) is a broader category of tool — AB 2575 defines it separately from "artificial intelligence," with AI as one possible component of a CDSS — that helps a clinician assess or decide something about a patient’s care. AB 2575 calls anything that triggers its rules a "covered tool."

As currently drafted, AB 2575 would require a health facility, clinic, physician's office, or group practice that uses a covered tool for patient care to give written notice to any licensed health care professional (or other person) who uses the tool or views its output. That notice would have to cover the tool's developer, funding source, and any foundation model used; its intended patient population, intended users, and intended decision-making role; known risks and out-of-scope uses; and — the operative override language — a statement that:

a worker providing direct patient care is permitted to override the output of a covered tool... if, in the judgment of the worker acting within their scope of practice, an override is appropriate for the patient, or as necessary to comply with applicable law, including civil rights law.

Timing matters here, and it’s worth stating precisely: the bill’s current text requires this notice at least 90 days before a new clinical decision support system is first deployed for patient care, plus notice to new staff at hire, onboarding, or credentialing, and an annual notice of their right to request the facility’s tool inventory. An earlier pass of this research briefly (and incorrectly) suggested this 90-day window had been removed by Senate amendment — a direct check of the current bill text confirms it is still there.

Would AB 2575 change who’s liable if an AI-drafted note causes harm?

This is the bill’s sharpest legal mechanic, and it’s more precise than "AI vendors can’t dodge liability." As currently drafted, the new Civil Code Section 1714.48 would state:

In an action against a defendant who developed, modified, selected, or deployed artificial intelligence or a clinical decision support system that is alleged to have caused harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the failure of a licensed health care professional or other health care worker to override an output of the artificial intelligence or clinical decision support system is a superseding cause severing the defendant’s liability for the alleged harm.

In plain terms: a vendor or facility that deployed the tool couldn’t point at the nurse who didn’t override a bad output and call that the whole cause of the harm. The bill explicitly preserves every other defense — The section does not limit or preclude a defendant from presenting any other affirmative defense, including evidence relevant to causation or foreseeability, or other evidence relevant to the comparative fault of any other person or entity. So this provision removes one specific argument, not liability itself.

Could your facility face a legal claim for disciplining a nurse who overrides an AI tool?

Separately from the disclosure duty, AB 2575 would add a new Labor Code article protecting staff who exercise that override right. As currently drafted, Labor Code Section 2821(b)(1) would state:

An employer shall not retaliate or discriminate against a worker providing direct patient care using their professional judgment to make an assessment or decision within their appropriate scope of practice based solely on the worker's override of, or reliance on, the output of a clinical decision support system.

Section 2821(c) would let that worker file a complaint directly with the Labor Commissioner. This is a distinct exposure from the liability bar above — an employment claim over disciplining a nurse for overriding a tool, not a patient-harm claim over the tool's output itself. A facility policy that pressures staff to defer to an AI tool's recommendation, or penalizes them for not doing so, is exactly the scenario this provision is aimed at.

Does AB 2575 apply to skilled nursing facilities specifically?

This is the question most facility administrators actually want answered, and the standing research brief this post is built on had originally missed it. Here’s the honest, careful version: AB 2575, as currently drafted, would reach SNFs because its own text borrows the existing Health & Safety Code Section 1250 definition of "health facility," which already includes skilled nursing facilities. The bill’s disclosure duty attaches to a "health facility, clinic, physician’s office, or office of a group practice," and AB 2575 defines "health facility" to have the same meaning as Health & Safety Code Section 1250. That section — itself enacted, current law, independently confirmed against California’s own statute mirrors — defines "health facility" to include the "skilled nursing facility" as one of its licensed categories (Health & Safety Code Section 1250, California Legislative Information; Justia mirror). No SNF exemption or carve-out was found anywhere in the bill text.

Read that carefully: this is a textual reading of what AB 2575 would do if enacted, not a statement that AB 2575 currently requires anything of your SNF, and not an independently adjudicated interpretation — it’s the strongest reading of the bill’s own cross-reference, checked once against the underlying statute, not yet tested by a court or confirmed by a second independent reviewer. Treat it as your starting point for asking your own counsel the right question, not as a settled compliance obligation.

One more honest gap: this reasoning is specific to facilities licensed as a "health facility" under Section 1250 — a category that covers SNFs. It does not extend, on the research behind this post, to RCFEs (Residential Care Facilities for the Elderly), which are licensed under separate statutory authority. If you operate an RCFE rather than a SNF, don’t assume this same analysis carries over automatically — that’s a genuinely open question this research didn’t resolve, and it’s worth raising directly with counsel rather than guessing either way.

When would AB 2575 take effect, and what happens next in the Legislature?

Multiple independent trackers — the Transparency Coalition’s weekly AI-legislation updates and the California Hospital Association’s own bill tracker — point to an Appropriations Committee action expected around early August 2026 as AB 2575’s next step, consistent with the standard "suspense file" process California uses for bills with a fiscal impact: they’re held in Appropriations and heard or released in a batch shortly before the mid-August fiscal deadline. Some secondary sources cite a specific August 3 date, but that couldn’t be confirmed against the Senate Appropriations Committee’s own posted agenda as of this writing, and it may reflect the full Legislature’s recess-return date rather than a hearing scheduled specifically for this bill. Treat "early August" as the right level of precision for now, and check back after the fact rather than planning around a specific calendar date.

If AB 2575 does clear Appropriations, passes the Senate floor, and is signed by the Governor, when would it actually take effect? California’s default rule — settled, current constitutional law — answers this directly. Article IV, Section 8(c) of the California Constitution provides that a statute enacted at a regular session shall go into effect on January 1 next following a 90-day period from the date of enactment of the statute, unless the statute carries an urgency clause, which takes a two-thirds vote in each house to attach. AB 2575’s confirmed Assembly floor vote was 48–15 — well short of the 54 votes an urgency clause would need. No source found in this research describes AB 2575 as carrying one. So if AB 2575 is enacted during the 2026 regular session, the current facts point to a January 1, 2027 effective date under the standard rule — a conditional projection from today’s vote count, not a confirmed date, since the bill hasn’t finished the legislative process yet.

What should your facility do while AB 2575 is still pending?

None of what follows is a legal requirement today — AB 2575 isn’t law yet. This is practical preparation, clearly separate from the factual claims above, for facilities that want a head start rather than a scramble if the bill is signed:

  • Inventory what you’d need to disclose. If you're already using or piloting an AI scribe or a clinical decision support tool, start listing its developer, any foundation model it uses, its intended use, and its known limitations — the same information AB 2575's disclosure duty would require, whether or not the bill passes.
  • Check your override culture, not just your policy. AB 2575's anti-retaliation provision assumes staff can actually exercise professional judgment against a tool's output without pushback. If your current workflow quietly discourages that, that's worth fixing regardless of what happens to this bill.
  • Ask your AI vendor how they'd support a 90-day notice. A vendor relationship that can't produce the disclosure content AB 2575 would require — developer identity, model details, risk disclosures — is a preparation gap worth surfacing now.
  • If you operate a SNF, ask your counsel the applicability question directly rather than waiting for guidance — the textual reading above gives you a specific, sourced starting point for that conversation.
  • Mark August 10, 2026 on your own calendar to check whether AB 2575 cleared Appropriations, and revisit this analysis before treating any part of it as settled.

This is also exactly the kind of audit trail a facility should already want independent of AB 2575: knowing who reviewed and signed off on an AI-drafted note, and being able to produce that record on demand. See how Notes Scribing handles that for long-term care documentation.

Frequently asked questions

Does AB 2575 apply to skilled nursing facilities?

On a textual reading, yes — AB 2575, as currently drafted, would reach SNFs because its own text borrows the existing Health & Safety Code Section 1250 definition of "health facility," which already includes skilled nursing facilities. But AB 2575 itself is not enacted law as of July 28, 2026, and this reading hasn’t been tested by a court or confirmed by more than one independent review — treat it as the starting point for a conversation with your own counsel, not a final answer.

Is AB 2575 law yet?

No. As of July 28, 2026, AB 2575 has passed the Assembly (48–15) and cleared two Senate policy committees, and is sitting in the Senate Appropriations Committee awaiting its next action, expected around early August 2026. It has not been signed, and everything in this post describes what the bill would do if enacted.

Would AB 2575 make our AI vendor automatically liable if something goes wrong?

No. As currently drafted, it removes one specific defense — a vendor or facility can't point to a clinician's failure to override the tool's output as the sole cause of harm. Every other defense, including comparative fault and causation arguments, is explicitly preserved in the bill's own text.

Can we discipline staff for overriding an AI tool's recommendation?

As currently drafted, AB 2575 would say no — a new Labor Code provision would bar retaliating against a direct-patient-care worker for overriding, or relying on, a covered tool's output within their scope of practice, with a right to file a complaint with the Labor Commissioner.

When would AB 2575 take effect if it's signed into law?

Based on its confirmed 48–15 Assembly vote (short of the two-thirds needed for an urgency clause), the standard constitutional rule would apply: January 1 following a 90-day period after enactment. For a bill signed during the 2026 regular session, that points to January 1, 2027 — a conditional projection based on today's facts, not a confirmed date, since the bill hasn't finished the legislative process.

Disclaimer: This post is informational, not legal advice. AB 2575 is pending legislation that has not been signed into law as of the "current as of" date above, and its provisions may change before final passage or fail to pass entirely. Confirm any compliance planning with your own counsel or state association before acting on it.

Sources

Where Relic Care fits in

If your facility is already using or piloting an AI scribe, the tool should be able to show who reviewed and signed off on each note, and produce that record on demand — the same kind of disclosure and audit trail AB 2575 would formalize if it becomes law. See how Notes Scribing handles that for long-term care documentation.

And if tracking bills like this one — alongside every other disclosure, consent, and retention obligation your facility carries — sounds like more spreadsheet than any one person has time for, that's what Compliance is built for.

Related reading: for the consent side of AI-scribe adoption — a separate legal question from AB 2575’s disclosure duty — see how patient consent for AI scribes already exposes facility liability today.

Read next

More for the People Running Your Facility