Texas changed the default in January. The Texas Responsible AI Governance Act (TRAIGA) requires every licensed healthcare provider in the state to give patients a written, visible disclosure when AI might be involved in their diagnosis or treatment. The sign goes up in the waiting room, the patient signs an acknowledgment, the EHR logs the disclosure event, and if the practice can't show its software vendor has documented its AI uses to the state, the practice gets a notice.
That was January. By June 2026, a similar law has been filed in 45 other states and DC. 240 bills, give or take, depending on how you count amendments. They don't all use the TRAIGA language — California's SB-833 takes a stricter line on automated decision-making in payer authorizations; New York's A.5675 requires real-time patient consent for any AI medical interpretation; Massachusetts's H.1973 extends the disclosure standard to nursing homes and assisted-living facilities. But the pattern is consistent enough that the legal community has started calling it the "TRAIGA cluster" — a thirty-year flashback to the medical informatics community calling HIPAA "the cluster" after it propagated through state implementation in the early 2000s.
For most families with a parent in the healthcare system, this is going to land first as paperwork. The waiver they signed at the cardiology visit in February will have one new clause by August. The portal will have a new banner explaining that the after-visit summary they're reading was AI-drafted and reviewed by a clinician. The voice on the practice's after-hours number will mention AI in its first sentence and offer to transfer to a human if requested.
A few of those changes are small. The cumulative weight of them — across one parent's cardiology + nephrology + primary care + specialist team across a year — is not small. And the absence of a national framework means each state's version looks slightly different in ways the average family caregiver isn't built to track.
This essay is the road map.
The four families of bills
Reading 240+ bills into one piece of writing requires categorization. The state-AI-in-healthcare bills cluster into four families, and the cluster a particular bill belongs to tells you what your family will experience under it:
1. Disclosure bills (most common; about 110 of the 240). Modeled on TRAIGA. Require a written, posted, signed disclosure whenever AI is used in patient diagnosis or treatment. Don't restrict the AI's use; just require the patient to know about it. Practices must train staff on the disclosure; software vendors must document their AI uses to state agencies. Examples: TX-TRAIGA (effective Jan 1, 2026), Illinois HB-3773, Ohio HB-211.
2. Decision-making bills (about 60 of the 240). Restrict the AI's authority to act, not just require disclosure. Examples: California SB-833 (prohibits AI from issuing prior-authorization denials in payer workflows; requires a licensed physician to make the call), Colorado HB-25-1212 (requires human-clinician sign-off on any AI-generated treatment recommendation that affects coverage), New Jersey A-4636 (the most restrictive — forbids any AI-driven decision in an emergency department without same-shift physician review).
3. Real-time consent bills (about 40). Require not just a posted disclosure but explicit patient consent at the moment the AI is invoked. Examples: NY A-5675 (real-time consent for AI medical interpretation — i.e., language translation by AI), Florida SB-602 (consent for AI in clinical-note dictation), Washington HB-1934 (consent for AI in chart pre-review before the visit).
4. Setting-extension bills (about 30). Apply existing or proposed disclosure rules beyond the typical "physician's office" to other care settings. Examples: MA H.1973 (extends disclosure to nursing homes and assisted-living), Pennsylvania HB-1421 (extends to home-health agencies), Minnesota SF-2841 (extends to school-based health services for kids with chronic conditions). This cluster has the most direct implications for family caregivers because nursing-home, home-health, and chronic-care-coordination contexts are exactly where families spend the most time.
The fragmentation is real and isn't going to consolidate quickly. State legislatures move at their own paces; the bills that pass will pass in different forms; the implementation rules each state's department of health publishes will diverge further. Families caring for a parent across state lines (a Florida snowbird whose primary care is in Massachusetts and whose summer specialist is in Wisconsin) are going to experience three different versions of the same rule.
What this means at the bedside
Concretely, here's what we expect to see in the next 12 months for a family caring for an aging parent across the four states the bills move first:
The waiver gets longer. Every new clinical relationship — new specialist, new lab, new hospital — comes with a disclosure-acknowledgment page. The family caregiver who handles paperwork for a parent will add this to the stack. The acknowledgment is signed once; the underlying disclosure may update without new signature in some states (Texas, Illinois) and require a new signature in others (New York, Florida).
The portal language shifts. Patient portals will start labeling AI-generated content explicitly. After-visit summaries will say "AI-drafted, reviewed by Dr. X." Suggested follow-up questions will say "AI-generated." Medication-interaction flags will say "AI-flagged, please discuss with your clinician." Most of these were already there; the labels are new.
Some workflows get slower. In states with real-time-consent bills (NY, FL, WA), some visits will pause for a one-line consent dialogue about AI use. The patient and clinician will trade a quick acknowledgment; the patient can refuse, but most won't (and shouldn't — refusal often means the human takes longer to do work the AI was helping with). The friction is small per visit; multiplied across a chronically-managed condition, it adds up.
Some workflows get safer. In states with decision-making bills (CA, CO, NJ), AI prior-authorization denials and AI emergency-department recommendations stop happening without a human in the loop. The "the algorithm said no, talk to your insurer" frustration drops. The trade-off: human-in-the-loop adds clinician time, which adds cost, which in some states will get passed back as longer waits.
The home-health setting gets the most scrutiny. Setting-extension bills (MA, PA, MN, others queuing) will require disclosure in home-health visits — the nurse who comes three days a week, the PT who comes on Tuesdays. Most home-health agencies use AI for scheduling, route optimization, and (increasingly) note-drafting. The disclosure will surface this to families who didn't know.
What families can do now
The AI-disclosure landscape is mostly the state's job to enforce. But families have two practical leverage points:
Ask the five questions. The questions we published earlier this week work in any state, under any policy regime. Whether your state has passed a disclosure law or not, you can ask: is AI being used in my parent's care, was it tested in patients like my parent, who reviews the output before it affects care, can I see what the AI saw, what happens if it's wrong. The good answers are the same; the bad answers tell you when to be cautious. The state laws are catching up to what thoughtful patients have been doing on their own.
Document the bilingual context. This is the most-overlooked compliance gap in the state-AI-bills cluster. Most of the bills require disclosure but say very little about disclosure in patient-preferred language. A Spanish-dominant grandmother in California whose family caregiver speaks English at the appointment will get an English disclosure; the AI use is disclosed to the family member, not to the patient herself. If your family's bilingual dynamic puts the patient in this position, raise it. Ask for the disclosure in their language. Title VI of the Civil Rights Act and Section 1557 of the ACA already require it; the new AI-disclosure laws add another reason to insist.
Keep records. If a state passes a disclosure law and the practice can't show it gave you one, that's a violation. The family that keeps a folder (or a shared workspace) with their parent's signed forms, after-visit summaries, and consent acknowledgments has the documentation needed to escalate if something goes wrong later. The folder lives in a binder, in a Google Drive, in a Kintaria document vault — wherever your family already keeps the picture. The point isn't the tool; it's that someone has the record.
What we're watching
Three things to watch in the next quarter:
Federal preemption. A few state bills (NY in particular) include language asserting the state can regulate AI in healthcare even if the federal government acts. The OMB and HHS have hinted at federal AI-disclosure guidance later in 2026; if it comes with explicit preemption, the state landscape simplifies. If it comes without preemption (more likely), the patchwork sticks.
The first enforcement action. Texas TRAIGA is six months old; the first complaint has been filed but no public enforcement action yet. Whoever the Texas AG names first will set the de facto template for which kinds of practices and which kinds of AI uses are the "first to be punished" pattern. Practices and software vendors watching for the signal are already adjusting.
The bilingual cases. The Spanish-dominant patient case described above is going to produce some early-test litigation — either a Section 1557 complaint or a direct state-AI-bill complaint. Whichever forum decides it first will set the precedent for how the disclosure-in-preferred-language requirement gets enforced. We'll be reading those filings.
For now: the state AI bills wave is real, it's about to land on every family with a parent in healthcare, and the leverage you have is mostly the same leverage you had before — asking good questions, documenting what you're told, and bringing language-access expectations into the conversation when they apply. The state laws are catching up to what good caregiving has always required.