California AB 3030 Requirements for Patient-Facing AI Messages
California AB 3030, codified as Health and Safety Code section 1339.75 and effective January 1, 2025, requires specified health care providers to disclose when generative AI is used to generate patient communications pertaining to clinical information. The statute also contains an exception tied to a care provider's involvement. A policy document does not show either was applied to a given message. Compliance teams need a control at the point of generation and sending, plus a per-message record.
Evaluating runtime enforcement against these requirements
- Does it decide per message, before delivery, or only detect afterward?
- Can it record reviewer identity and decision for each escalated message?
- Can disclosure checks vary by written, audio, and video formats?
- Is evidence exportable per message, with PHI minimized and access restricted?
- Can retention and data residency be configured to your policy?
- Can policy changes be versioned so each record shows the rule applied?
What AB 3030 establishes
AB 3030 (2023-2024 session) is titled Health care services: artificial intelligence. It added Chapter 2.13, Artificial Intelligence in Health Care Services, to Division 2 of the Health and Safety Code. The chapter appears to contain a single section, 1339.75, which was added by Stats. 2024, Ch. 848 and took effect January 1, 2025. As of November 2026, no amendment is reported.
The Governor's office describes the core obligation as follows: specified health care providers must disclose the use of generative AI when it is used to generate communications to a patient pertaining to patient clinical information. That sentence contains two tests a system can evaluate for every message. The first is whether the content was generated by generative AI. The second is whether it pertains to clinical information.
The second test is where most programs need a decision rule. A message inventory should separate communications about clinical information from administrative ones, and borderline cases, such as a reminder that mentions a test result, need a documented classification outcome. Counsel should confirm the exact statutory scope before those rules are fixed.
The care-provider exception
Section 1339.75 includes an exception under which the requirements of subdivision (a) do not apply when a condition involving a care provider is met. For compliance purposes, the exception matters because it changes what the system must prove. If the disclosure path applies, the record must show the disclosure was present. If the exception path applies, the record must show that the statutory condition was satisfied for that specific message.
That means review status has to be a recorded fact, not an assumption. A workflow flag such as "reviewed" should be defined from the statute's exact wording as read by counsel, and it should be set only by an identifiable reviewer action before the message is sent. A general statement that clinicians oversee AI output will not demonstrate that a particular message qualified.
There is an operational tradeoff. Routing messages to provider review adds latency and clinician workload. Applying disclosure scales more easily but must be verified on every message. Many organizations will use both paths and decide per message type which one applies.
Mapping each requirement to a control point and evidence
The table of obligations below is a design approach, not a statutory checklist. Each row pairs an obligation with the place in the generation and delivery path where it can be enforced, and with the artifact that shows it happened. Disclosure wording and format-specific placement for written, audio, and video messages should be held as policy parameters set from counsel's reading of the statute.
- GenAI use identified: Control: tag the message as GenAI-drafted at generation. Evidence: the tag, with message identifier and timestamp.
- Clinical-information scope: Control: classify the message before send using the inventory's decision rules. Evidence: classification outcome and the rule version applied.
- Disclosure applied: Control: pre-send check that the required disclosure is present in the format being delivered. Evidence: pass or fail result per message.
- Provider review exception: Control: hold the message until a reviewer acts. Evidence: reviewer identity, time, and decision.
- Non-compliant message: Control: block or escalate instead of sending. Evidence: the enforcement decision and its reason.
- Delivery: Control: format-specific handling at the channel. Evidence: delivery time and channel.
Evidence, PHI, and HIPAA audit controls
A regulator or auditor reviewing AB 3030 compliance is likely to ask about specific messages, not about policy. A useful per-message record contains the message identifier, the classification result, the GenAI tag, the policy version in force, whether disclosure was applied or review was recorded, the reviewer where relevant, the enforcement decision, and the delivery details. The statute as summarized here does not prescribe a log format or retention period, so those choices belong to your counsel and records policy.
HIPAA adds a second requirement. Section 164.312(b) calls for mechanisms that record and examine activity in information systems that contain or use electronic PHI. A per-message record produced by enforcement is one way to support that audit control. The record itself can contain PHI, so apply minimum-necessary design: store references or redacted content where the decision does not need the full text, restrict access to the evidence store, and set retention deliberately.
This guide is not legal advice. Mapping evidence to HIPAA and to AB 3030 should be reviewed by counsel.
Who is accountable and how enforcement works
Enforcement routes differ by entity type. A violation of section 1339.75 by a licensed health facility is subject to the enforcement mechanisms in Article 3 (commencing with Section 1275) of Chapter 2. For a licensed clinic, the mechanisms are those in Article 3 (commencing with Section 1225) of Chapter 1. For a physician, jurisdiction sits with the Medical Board of California or the Osteopathic Medical Board of California, as appropriate. Identify which of your entities fall into each category and assign accountability accordingly.
AB 3030 is not the only obligation. The California Attorney General's January 2025 advisories state that entities must follow existing California laws alongside new laws effective January 1, 2025. For healthcare, that includes consumer protection, civil rights, data privacy, and professional licensing law. The Attorney General also notes the advisories are not exhaustive. Treat AB 3030 controls as one component of a broader healthcare AI compliance program.
AB 3030 as a per-message control path
Classify
Is the message GenAI-drafted and does it pertain to clinical information?
Route
Apply disclosure, or send to a care provider for review.
Enforce
Allow, redact, block, or escalate before delivery.
Evidence
Retain the decision record for each message.
Turn AB 3030 into evidence you can produce per message
Talk through how enforcement in the execution path can route patient messages, apply policy before delivery, and generate audit-ready evidence.
Talk to an Expert