criticalaccess.aiAI for rural hospitals

Regulatory map · Last checked November 18, 2026

The rules on AI in health care, sorted by who they bind.

Most of these rules were not written for a 25-bed hospital. Some bind you directly. Others bind your EHR vendor or the health plans that pay you, and matter because of what you can ask for. Each entry says which.

General information, not legal advice. Laws change, and several entries below changed in the past year. Confirm with the linked source and your counsel before you act.

How to read the map

Every entry carries a status and a source.

Dates come from the rule, the statute, or the agency’s own page wherever we could read it. Where we relied on published legal trackers, the entry says so.

  • In force Applies today
  • Dated Takes effect or changes on a set date
  • Proposed Not final, or guidance only
  • Voluntary No one is required to follow it
  • Tracker-sourced Dates from legal trackers. Check the statute.
A note on terms

An executive order signed September 29, 2026 directs federal agencies to write “Super Intelligence” or “SI” where they used to write “artificial intelligence” or “AI.” The order defines the new term by the existing legal definition of AI, so it covers the same technology. FDA began using it on November 16, 2026, and older documents stay as written.

If you search an agency site and find “SI,” it is the same subject. This site keeps saying AI.

01 · Federal

Rules that bind the hospital.

Two federal rules apply to your own use of AI today. Neither mentions critical access hospitals, and neither exempts them.

In force

Nondiscrimination in decision support tools

HHS Office for Civil Rights · 45 CFR 92.210

General rule since July 5, 2024. Duty to identify and mitigate since May 1, 2025.

What it requires

A hospital that takes federal funds must make reasonable efforts to identify the patient care decision support tools it uses that take race, color, national origin, sex, age, or disability as inputs, and to reduce the risk of discrimination from each. It covers any clinical algorithm, whether or not anyone calls it AI.

What to do

Keep a list of those tools, including the ones built into your EHR. For each, write down what the vendor says about those inputs and how you check the results. That list is also the start of your AI inventory.

Where it stands

A federal court vacated other parts of the 2024 rule in October 2025. HHS confirmed in a June 2, 2026 notice that the remaining provisions are still in force.

Sources: final rule, 89 FR 37522 · text of § 92.210 · HHS notice, June 2, 2026

In forceUpdate proposed

HIPAA Security Rule

HHS Office for Civil Rights · 45 CFR Part 164

Update proposed January 6, 2025. Not final at this review.

What it requires

The current rule applies to any AI tool that touches patient information, the same as any other system. The proposed update would add specifics, including a technology asset inventory, multi-factor authentication, and encryption.

What to do

Sign a business associate agreement before any vendor’s AI touches patient data. That includes AI features switched on inside tools you already use. Keep patient information out of general chatbots that have no agreement.

Source: proposed rule, 90 FR 898

What we did not find

A federal rule written for ambient documentation tools. At this review, the federal rules that apply to an AI scribe are the general ones above. New duties to tell patients are coming from the states: see Texas, Louisiana, and Rhode Island.

02 · Federal

Rules that bind your vendors.

These rules decide what a vendor has to show you. Both moved toward less oversight in 2026, so more now depends on what your contract says.

Changes Dec. 28, 2026

EHR certification: transparency for predictive tools

ASTP/ONC · 45 CFR 170.315(b)(11)

Required since January 1, 2025. A final rule published October 28, 2026 removes it, effective December 28, 2026.

What it requires

Developers of certified EHRs have had to show “source attributes” for the predictive decision support tools they supply: plain facts about how each one was built, tested, and meant to be used. The new rule removes the source attribute and risk management requirements. A narrower decision support requirement remains.

What to do

Ask your EHR vendor now, in writing, for the source attribute information on every predictive tool in your system, and whether they will keep providing it. After December 28 that depends on your contract. The certification rule never covered AI tools you buy from anyone else.

Sources: 2024 rule (HTI-1), 89 FR 1192 · 2026 final rule (HTI-5), 91 FR 67916

Guidance

FDA: decision support and wellness software

Food and Drug Administration

Revised final guidance, January 2026. A January 2025 draft on AI-enabled devices is still a draft.

What it says

FDA replaced its 2022 guidance on clinical decision support software and its 2019 guidance on general wellness products. Legal analyses of the revisions describe them as widening the software and consumer devices that FDA does not review as medical devices.

What to do

Ask every vendor whether the product is an FDA-authorized device, and get the answer in writing. “FDA doesn’t regulate it” is not a safety review, so tools outside FDA’s reach belong in your own governance. FDA keeps a public list of the AI-enabled devices it has authorized.

Sources: decision support guidance · general wellness guidance · FDA device list · analysis by Arnold & Porter

03 · Federal

Rules that bind health plans.

You don’t have to comply with these. Your payers do, and they belong in your denial appeals.

In force

Medicare Advantage: algorithms in coverage decisions

CMS · 42 CFR 422.101(b)–(c)

Applies to coverage since January 1, 2024. CMS answered questions on AI in a memo dated February 6, 2024.

What it requires

A Medicare Advantage plan may use an algorithm to help, but a coverage decision has to rest on the individual patient’s circumstances. CMS’s memo says “algorithms or artificial intelligence alone cannot be used as the basis to deny admission or downgrade to an observation stay,” and that a predicted length of stay alone cannot end post-acute care.

What to do

Put the regulation and the memo in your appeal templates. When a denial reads like a formula, ask the plan in writing whether an algorithm was used and who reviewed the case.

Where it stands

CMS proposed more specific limits on AI for 2026 and did not finalize them.

Sources: final rule, 88 FR 22120 · CMS memo of February 6, 2024 (copy hosted by AHA)

Live in six states

WISeR: technology-assisted prior authorization in Original Medicare

CMS Innovation Center

January 1, 2026 through December 31, 2031. Applies to dates of service from January 15, 2026.

What it does

In Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington, selected outpatient services in Original Medicare need prior authorization, handled by a technology vendor working with the Medicare contractor. Included services billed without it go to review before payment.

What to do

In those six states, check CMS’s list of included services against what you bill, and learn your state vendor’s process. Elsewhere, watch for an expansion.

Where it stands

The Government Accountability Office found in May 2026 that the model is a rule subject to congressional review, and members of Congress have moved to block it. It remained in effect at this review.

Sources: CMS model page · CMS questions and answers · GAO decision B-337994

04 · Federal and national

Voluntary frameworks, funding, and what to watch.

None of these is a requirement today. One is a free outline for your AI policy, one is money, and one is a fight over who gets to make the rules.

Voluntary

Joint Commission and CHAI: responsible use of AI

The Joint Commission · Coalition for Health AI

Guidance, September 2025. Governance playbooks, May 27, 2026. Voluntary certification since June 1, 2026.

What it says

The guidance names seven elements: policies and governance, patient privacy and transparency, data security, ongoing quality monitoring, voluntary blinded reporting of safety events, risk and bias assessment, and education and training. None of it is an accreditation standard.

What to do

Use the seven elements as the outline for your own AI policy. The certification is open to critical access hospitals and does not require Joint Commission accreditation. Pursue it only if you want an outside review.

Sources: guidance document · CHAI playbooks announcement

Year two awarded

Rural Health Transformation Program

CMS · Public Law 119-21, section 71401

Year one awards December 29, 2025. Year two awards November 9, 2026.

What it is

$50 billion to states over five federal fiscal years, 2026 through 2030. The uses named in the law include “training and technical assistance for technology-enabled care delivery in rural hospitals, including remote monitoring, robotics, and artificial intelligence,” and technology advances including cybersecurity.

What to do

CMS does not pay hospitals directly. Find your state’s program and its subaward rounds. Cybersecurity, telehealth, and training for technology-enabled care are named uses, so those are the budget lines to ask about.

Sources: CMS program page · year one awards · year two awards

Watch

Federal challenge to state AI laws

Executive Order 14365

Signed December 11, 2025.

What it does

The order directs federal agencies to challenge state AI laws it considers burdensome and asks Congress for a national framework. The Justice Department set up a litigation task force in January 2026.

What to do

Follow your state’s law as written. At this review we found no federal action that has set aside a state law on AI in health care, and no preemption statute has passed.

Sources: Executive Order 14365 · Attorney General memorandum

05 · State laws

Where the new duties are appearing.

State laws fall into three groups: telling patients when AI is used, limiting how payers use AI to deny care, and restricting AI in mental health care. A state not listed here had no health care AI law we could confirm.

This list is not complete, and the 2026 entries marked tracker-sourced have not been checked against the statute.

AlabamaPayers

SB 63. Insurers must disclose their use of AI in prior authorization, base AI-influenced decisions on the individual’s history, and have denials made by a licensed physician or professional. Effective October 1, 2026.

Source: Holland & Knight state roundup

Tracker-sourced

ArizonaPayers

HB 2175. A medical director must personally review any denial of a claim or prior authorization on medical necessity, and may not rely solely on another source. The enacted text does not use the term AI. Effective July 1, 2026.

Source: Arizona Legislature bill summary

In force

CaliforniaDisclosure · Payers · Clinical use

AB 3030 (since January 1, 2025). A facility, clinic, or physician office that uses generative AI to write patient communications about clinical information must say so and tell the patient how to reach a person, unless a licensed provider reviewed the message.

SB 1120 (since January 1, 2025). Health plans may not use AI as the sole basis to deny, delay, or modify care. A licensed clinician makes medical necessity decisions.

AB 489 (since January 1, 2026). AI may not use titles or wording that imply a licensed health professional is giving the advice.

AB 1979 and SB 503 (from January 1, 2027). A licensed provider must exercise independent judgment when decision support software informs care, and developers and users of that software must work to identify and reduce biased results.

Sources: Department of Insurance guidance on SB 1120 · California Health Care Foundation on the 2026 bills · Duane Morris on AB 3030

In force

ColoradoClinical use · Payers · Mental health

SB 26-189 replaced the state’s 2024 AI Act before it took effect. From January 1, 2027, an organization using automated decision technology in a consequential decision, which includes health care services, must give notice, explain its role after an adverse decision, and offer human review. A federal court order has put enforcement on hold while a lawsuit proceeds.

HB 26-1139 (from January 1, 2027). Payers may not deny care on medical necessity based solely on AI output, and a licensed clinician must review.

HB 26-1195 (since August 12, 2026) prohibits AI therapy chatbots and limits AI use by licensed psychotherapy providers.

Sources: SB 26-189 · HB 26-1139 · Norton Rose Fulbright analysis

Jan. 1, 2027

GeorgiaPayers · Chatbots

SB 444 (from January 1, 2027). Utilization review AI may not issue an adverse determination until a person conducts a review with a clinical peer. See the Georgia section below.

Jan. 1, 2027

IllinoisMental health · Payers

HB 1806 (since August 1, 2025). Therapy may not be delivered by AI. Licensed professionals may use AI for administrative tasks, but not to make therapeutic decisions or talk with clients in their place.

SB 3114 (from January 1, 2028). Payers may not use automated tools to downcode a claim without a person reviewing what the billing professional submitted.

Sources: HB 1806 status · Public Act 104-0568

In force

IndianaPayers · Billing

HB 1271. AI may not be the sole basis for downcoding a claim without review by a health professional. Providers may not submit AI-generated claims without professional review. Effective July 1, 2026.

Source: Holland & Knight state roundup

Tracker-sourced

IowaPayers

HF 2635. Insurers may use AI for a first review of prior authorization but may not deny, delay, or downgrade care based on AI alone. Effective July 1, 2026.

Source: Transparency Coalition roundup

Tracker-sourced

LouisianaDisclosure

Act 649. A licensed health professional must tell the patient, out loud, before recording any part of an appointment that AI will transcribe. The law requires disclosure, not consent. Effective August 1, 2026.

Source: Arnall Golden Gregory analysis

Tracker-sourced

MaineMental health

LD 2082. Only licensed professionals may provide therapy. They may use AI for administrative and limited tasks, and need the patient’s consent before ambient listening or AI recording. Effective July 29, 2026.

Source: Holland & Knight state roundup

Tracker-sourced

MarylandPayers

HB 820 (since October 1, 2025). Carriers and review agents using AI in utilization review must base decisions on the patient’s own clinical history, not group data alone, and report quarterly on AI use in adverse decisions.

Source: Maryland General Assembly, HB 820

In force

NebraskaPayers

LB 77. An AI algorithm may not be the sole basis to deny, delay, or modify care on medical necessity. Insurers must disclose AI use to the state, to providers, and to enrollees. Applies to plans issued or renewed on or after January 1, 2026.

Source: Nebraska Legislature, LB 77

In force

NevadaMental health

AB 406 (since July 1, 2025). An AI system may not be offered as providing professional mental or behavioral health care. Providers may use AI for administrative tasks such as scheduling and billing.

Source: Wilson Sonsini analysis

In force

OregonTitles

HB 2748 (since January 1, 2026). A nonhuman entity, including an AI agent, may not use the title nurse or any nursing license title.

Source: Oregon Laws 2025, chapter 378

In force

Rhode IslandDisclosure · Mental health

Use of AI by Healthcare Providers Notification Act (since June 22, 2026). Licensed providers and facilities must notify patients when AI documents an in-person or telehealth visit. Patients may opt out, and the provider must review the documentation for accuracy.

Sources: Rhode Island General Assembly release · Nixon Peabody analysis

In force

TennesseeMental health

SB 1580. AI may not be advertised or represented as a qualified mental health professional. Effective July 1, 2026.

Source: Holland & Knight state roundup

Tracker-sourced

TexasDisclosure · Payers

HB 149 (since January 1, 2026). When AI is used in relation to a health care service or treatment, the provider must disclose that to the patient no later than the date the service is first provided, or as soon as reasonably possible in an emergency. The disclosure must be clear and in plain language, and may be a link to a separate web page.

SB 1188 (since September 1, 2025). Practitioners who use AI for diagnostic purposes must disclose it to patients and review AI-generated records.

SB 815 (since September 1, 2025). AI may not make an adverse medical necessity determination in the health benefit claims process.

Sources: HB 149, enrolled text · SB 1188 · SB 815

In force

UtahDisclosure · Mental health · Payers

Artificial Intelligence Policy Act (2024, amended May 7, 2025). People in licensed occupations, including health professions, must disclose generative AI in high-risk interactions such as giving medical or mental health advice.

HB 452 (since May 7, 2025) sets disclosure and data rules for mental health chatbots. SB 319 (from January 1, 2027) requires insurers to disclose AI use in prior authorization.

Source: Perkins Coie analysis

In force

VermontMental health

Act 156. AI may not independently deliver mental health services. Signed June 17, 2026.

Source: Transparency Coalition roundup

Tracker-sourced

WashingtonPayers

SB 5395. Only a licensed physician or professional may deny care on medical necessity. AI may not be the sole means to deny, delay, or modify care, and AI-aided denials are reported to the insurance commissioner. Effective June 11, 2026.

Source: Holland & Knight state roundup

Tracker-sourced
What this means for a rural hospital

Revenue cycle: in the states above with payer laws, a medical necessity denial that rests on AI alone can be appealed as out of compliance, and several states make payers disclose AI use, which gives you grounds to ask.

Clinical operations: Texas, Louisiana, and Rhode Island now require telling patients before or when AI documents a visit. A short, plain notice is cheap to adopt everywhere, whether or not your state requires it yet. The ambient documentation playbook covers how.

06 · Georgia

One payer law, one chatbot law, and state funding.

This site is written in Georgia, so Georgia gets a closer look. Other states will follow as hospitals ask.

Starts Jan. 1, 2027

AI in utilization review

SB 444, Act 411 · O.C.G.A. § 33-46-7.1

Signed May 5, 2026. Effective January 1, 2027.

What it requires

Review agents and utilization review entities may use AI tools, but a tool may not issue an adverse determination until a person conducts a review in which a clinical peer takes part. The tool may never supersede that clinical peer’s judgment. The law does not require telling patients about AI.

What to do

From January 2027, cite the new code section when you appeal an adverse determination from a Georgia-regulated review entity, and ask who the clinical peer was.

Source: SB 444 as passed, Georgia General Assembly

Starts July 1, 2027Tracker-sourced

Conversational AI Safety Act

SB 540

Signed May 2026. Effective July 1, 2027.

What it requires

Conversational AI must disclose that it is AI, with added protections for minors. It is not a health care law, but a patient-facing chatbot on a hospital website would fall under the disclosure duty.

What to do

If you run or plan a chatbot for patients, make sure it says plainly that it is not a person.

Source: Georgia Senate press office

Open rounds

GREAT Health: Georgia’s rural transformation funding

Georgia Department of Community Health

Federal awards to Georgia: $218,862,170 for year one and $221,731,766 for year two.

What it is

Georgia runs its share of the Rural Health Transformation Program as a subaward program. Its second round, announced July 16, 2026, awarded $30.6 million to 26 organizations, including funding for rural hospital cybersecurity.

What to do

Watch the state’s announcements for the next application round, and have a short, specific technology request ready.

Source: Georgia DCH announcement

What we did not find in Georgia

A state law, regulation, or licensing board rule on a hospital’s or physician’s own use of clinical AI, or a state duty to tell patients about AI documentation tools. That is an absence in our search, not an official statement. Federal law and ordinary professional standards still apply.

07 · Coming up

Dates already on the calendar.

Dec. 28, 2026
Federal. The EHR certification requirement to show source attributes for predictive tools ends.
Jan. 1, 2027
Georgia SB 444. Colorado SB 26-189 duties and HB 26-1139. California AB 1979 and SB 503. Utah SB 319.
July 1, 2027
Georgia SB 540, the chatbot disclosure law.
Jan. 1, 2028
Illinois SB 3114, on automated downcoding of claims.

This quarterFive things worth doing now

  • List your decision support tools, including the ones inside the EHR. One list serves Section 1557 and your AI inventory.
  • Ask your EHR vendor in writing for source attribute information before December 28, and whether it will continue.
  • Check every agreement. No AI feature touches patient data without a business associate agreement.
  • Add a plain patient notice for AI documentation, whether or not your state requires one yet.
  • Update your appeal templates with the Medicare Advantage rule and your state’s payer law.

08 · What changed

The past year, newest first.

A running log of the changes that moved an entry on this map.

  • Federal

    Year two Rural Health Transformation awards announced

    $10 billion to all 50 states for the program’s second year.

    Nov. 9, 2026
  • Federal

    EHR certification rule drops source attribute requirements

    The HTI-5 final rule was published, effective December 28, 2026.

    Oct. 28, 2026
  • State

    California signs AB 1979 and SB 503

    Licensed provider judgment over decision support, and bias mitigation duties, from January 1, 2027.

    Sept. 30, 2026
  • Federal

    Agencies directed to use the term “Super Intelligence”

    Executive Order 14434. Same technology, new name in federal documents.

    Sept. 29, 2026
  • State

    Rhode Island requires notice when AI documents a visit

    Effective on signing. Patients may opt out.

    June 22, 2026
  • Federal

    HHS confirms the Section 1557 rule remains in force

    Apart from the provisions a court vacated in 2025.

    June 2, 2026
  • National

    Joint Commission opens a voluntary AI certification

    Open to critical access hospitals. Not an accreditation standard.

    June 1, 2026
  • State

    Colorado replaces its AI Act

    SB 26-189, with duties starting January 1, 2027.

    May 14, 2026
  • State

    Georgia signs SB 444 on AI in utilization review

    Effective January 1, 2027.

    May 5, 2026
  • Federal

    FDA revises its decision support and wellness guidance

    Replacing the 2022 and 2019 versions.

    Jan. 2026
  • Federal

    WISeR begins in six states, and Texas HB 149 takes effect

    Prior authorization in Original Medicare, and a patient disclosure duty for Texas providers.

    Jan. 1, 2026

See something out of date?

Rules move faster than any page. If an entry is wrong or your state is missing, send the citation and it will be checked and added.