What the law says
AB 1979 (Bonta) was chaptered on September 30, 2026 as Chapter 854 of the Statutes of 2026 and takes effect January 1, 2027 (CalMatters Digital Democracy). It does two separate things.
It brings health chatbots under California's medical privacy law. The bill adds a definition to the Confidentiality of Medical Information Act (CMIA). A "health care chatbot" is a generative AI system with a natural-language interface that gives adaptive, human-like responses, is marketed as facilitating or supporting health services to a consumer, and uses "health care chatbot information" to do it. That information is broad: anything about a consumer's physical or mental health or wellness that the consumer gives the chatbot, directly or by allowing access, plus anything the chatbot collects, generates or infers. A business that offers such a chatbot to help a person manage their information, or to diagnose, treat or manage a medical condition, is deemed a provider of health care for CMIA purposes. It must keep the same confidentiality standards as a provider and is subject to the CMIA's penalties for improper use or disclosure. The designation applies to the CMIA only, not to other laws governing providers.
It limits what AI may decide in clinical care. A new chapter of the Business and Professions Code, starting at section 22758.5, requires health facilities, clinics, physician offices and group practices to take reasonable steps so a licensed provider exercises independent professional judgment whenever a clinical decision support system informs care. They may not use AI to independently perform a clinical function the law reserves for a licensed person, or to direct unlicensed staff to perform one. Tools for scheduling, reminders, patient education, payment processing and routine communication that involves no professional judgment are outside it. Licensing boards enforce it, including the Medical Board of California (Holland & Knight; CHCF).
Our summary is based on the chaptered text as reported by those sources. Read the text itself on leginfo before you rely on any single phrase. This is not legal advice.
Why the privacy half matters more than it looks
Many telehealth brands assumed the CMIA did not reach their marketing layer. Since 2014, Civil Code § 56.06 has treated a business that offers consumers software or hardware designed to maintain medical information as a provider for CMIA purposes (Civil Code § 56.06). AB 1979 removes the argument for chatbots specifically, and its definition covers wellness information and inferences, not only medical records.
The CMIA has teeth. Section 56.36 lets a patient recover nominal damages of $1,000 for a negligent release of medical information without proving actual harm, plus actual damages, and adds administrative and civil penalties (Civil Code § 56.36). Multiply by every conversation your chatbot logged and the exposure is not theoretical.
In practice it means three things for a chatbot that talks to California patients about their health:
- What a patient types into it is medical information. It should not reach an ad platform, a marketing analytics tool or a model vendor's training set without the patient's authorization.
- The chat widget is a tracking surface. If your site chatbot loads on a page with a Meta pixel, check what that pixel sees. Our FTC v. Hims post covers pixel hygiene.
- Chat transcripts need the same access controls, retention rules and vendor contracts as the chart.
The other California AI laws, side by side
| Law | In force | Who it covers | What it requires |
|---|---|---|---|
| AB 3030 | January 1, 2025 | Health facilities, clinics, physician offices and group practices using generative AI for patient communications about clinical information | A disclaimer that the message was AI-generated (at the start of written messages) and clear instructions for reaching a human provider; exempt if a licensed or certified provider reads and reviews it first |
| AB 1979 | January 1, 2027 | Businesses offering consumer health chatbots (CMIA); facilities and practices using clinical decision support | CMIA confidentiality and penalties; a licensed clinician's independent judgment over AI |
| SB 503 | January 1, 2027 | Developers and deployers of clinical decision support | Identify and mitigate risks of biased impacts; developers give deployers documentation; deployers monitor |
| AB 1609 | January 1, 2027 | Businesses with over $500 million in annual revenue | Disclose a customer service bot, make a good-faith effort to connect to a human within 15 minutes of a request in business hours, post contact information; fines up to $5,000 and $10,000 |
Sources: Arent Fox Schiff on AB 3030; Holland & Knight and CHCF above. Governor Newsom vetoed SB 903, which would have restricted AI in psychotherapy, and AB 2575 on clinician override protections.
Most founder-led brands are far below AB 1609's $500 million threshold. AB 3030 and AB 1979 do not have thresholds.
What Ro's Robin shows
On October 29, Ro introduced Robin, an AI agent in its app that draws on the medications Ro prescribes, labs its clinicians order, logged side effects and provider counseling notes. When a patient reports a side effect, Robin routes them to a provider and drafts a summary the patient can edit, according to coverage of CEO Zachariah Reitano's Business Insider interview (AIM Media House). Ro's AI Agent Acceptable Use Policy, as quoted there, says the agent gives "general educational information only," does not diagnose or prescribe, is not monitored in real time, and that its answers are not reviewed by a person before the patient sees them.
We take no view on whether Robin complies with anything. The useful part is that Ro wrote down, in plain words, what its agent is and is not. Every brand with an AI touchpoint needs that paragraph by January.
Your AI touchpoints, one by one
| Touchpoint | Likely treatment in California from January 1 | What to change |
|---|---|---|
| Website chatbot that answers "am I eligible for a GLP-1?" | A health care chatbot under AB 1979; CMIA applies | No ad or analytics tags on the chat; vendor contract barring training on transcripts; transcripts stored and retained like chart data |
| AI support agent in the portal answering order and billing questions | Mostly administrative; becomes a health care chatbot as soon as it discusses symptoms or medication | Hard handoff to a human on anything clinical, and say so; disclose that it is AI at first contact |
| AI that drafts clinical messages a clinician sends | AB 3030 exemption applies if a licensed provider reads and reviews each message | Log the review; do not let unreviewed drafts go out on a timer |
| AI that sends clinical messages with no review | AB 3030 disclaimer and human-contact instructions required (for covered practices) | Add both, or put a clinician in the loop |
| Intake logic or AI that recommends eligibility, dose or product | Clinical decision support under AB 1979 when used by a covered practice | The clinician decides and the chart shows it; no auto-approval; no product preselected for the patient |
| Scheduling and refill reminder bots | Outside AB 1979's clinical rules | Keep them on logistics; the CMIA still applies to any health data they hold |
The fourth and fifth rows are where we see the most risk in DTC funnels. A "you qualify for semaglutide 0.5 mg, check out now" screen generated by software, before a clinician has looked, is the pattern these laws are written against. The clinician should decide eligibility and dose, and the record should show that they did.
A pre-January checklist
- List every place a patient can type to something that answers. Include the vendors behind each.
- For each, write down where transcripts go, who can read them, how long they are kept, and whether the vendor can train on them. Fix anything that sends health text to marketing tools.
- Add a first-message disclosure to every AI agent: it is AI, what it can help with, what it cannot, and how to reach a person. Our consent and terms post covers where similar language lives in your legal documents.
- Draw the clinical line in the agent's instructions and test it. Ask it about a side effect, a missed dose and a dose increase, and confirm each goes to a human.
- Have your medical director sign a one-page note on how AI is used in clinical workflows and who reviews what. That note is your evidence of "independent professional judgment."
- If you use any AI eligibility or triage logic, ask the vendor for its documentation on testing and intended use. SB 503 makes that a developer duty from January.
Where Turbopills fits
Our support model has been AI-first with human escalation since we wrote it up in February: an agent handles routine order, billing and refill questions from live data, and clinical questions go to your care team. Eligibility and prescribing stay with licensed providers through state-aware routing and provider review; intake collects answers and structures them for the clinician, it does not approve anyone. Agentic support in production covers the operating side.
What stays with you: your medical director's protocols, your disclosures and privacy notice, and the decision on which AI features to switch on for California patients. We are in private beta. In a demo, ask us where chat transcripts are stored, who can see them, and how the handoff to a human looks from the patient's side.
FAQ
What is California AB 1979? A 2026 law, effective January 1, 2027, that treats businesses offering consumer generative AI health chatbots as providers of health care under the Confidentiality of Medical Information Act, and requires clinics and physician practices to keep a licensed clinician's independent judgment in charge whenever AI clinical decision support informs care.
Does AB 1979 apply to telehealth companies outside California? It applies to businesses offering covered chatbots to consumers, and the CMIA protects California patients' information. A brand based elsewhere that serves California patients should assume it is in scope and confirm with counsel.
Do I have to tell patients they are talking to AI in California? For covered practices, AB 3030 has required a disclaimer and human-contact instructions on AI-generated clinical communications since 2025, unless a licensed provider reviews them first. AB 1609 adds a bot-disclosure duty in customer service for businesses over $500 million in revenue. Disclosing at first contact is the safe default for everyone.
Can an AI chatbot approve a patient for a prescription? No. Under AB 1979, covered practices may not use AI to independently perform clinical functions reserved for licensed professionals. A licensed clinician must make and document the decision.