What changes on Jan. 1

Laws in at least eight states that touch AI in health care take effect on Jan. 1. Most bind insurers and chatbot companies. A few, in California and Colorado, reach the practice itself.

Illustration of an hourglass with its last grains of sand falling, the sand picked out in amber.

From Jan. 1, a California physician's office may not use artificial intelligence to direct unlicensed staff in clinical work that the law reserves for licensed professionals, or to do that work itself. The law, signed by Gov. Gavin Newsom on Sept. 30, also requires health facilities, clinics, physicians' offices and group practices to take reasonable steps so that a licensed clinician keeps the ability to exercise independent professional judgment whenever the output of clinical decision support software informs a patient's care.

The same day, laws bearing on AI in health care take effect in at least eight states. Most of their duties fall on insurers and on companies that run chatbots. Only a few reach the practice itself, chiefly in California and Colorado. They come in four kinds: rules for clinical software, notice to patients, a person's review of insurers' denials, and limits on chatbots.

California defines clinical decision support as AI that produces a prediction, classification, recommendation, evaluation or analysis to help decide the timing of care, a diagnosis or a treatment. The law leaves alone AI used for documentation and communication that involves no professional judgment, and violations are subject to enforcement by the Medical Board and other licensing boards. A second law, SB 503, requires the companies that build such software and the providers that use it to make reasonable efforts to find tools with a known or foreseeable risk of biased results, meaning worse access, care or outcomes by race, sex, age or another protected trait. For the tools identified as carrying that risk, the providers must monitor them regularly and take reasonable steps to reduce it, and developers must give them documentation on training data and on how each tool was tested for performance and bias.

Colorado's new law on automated decision-making exempts health care providers covered by HIPAA, and their business associates, from most of its duties. A covered provider operating from a Colorado location must still give patients a general notice of its use of advanced technologies, which it may fold into other notices describing patients' rights and how it provides care. If that technology is used to decide whether a patient qualifies for financial assistance, the provider must explain its role, the information it relied on and how to ask for a correction or a human review. Decisions about employment are not exempt. In Utah, an insurer that uses AI to review requests for prior authorization must say so on its website and tell the state insurance department, each enrollee and each health care provider in its network.

Four states require a person, not software alone, to stand behind an insurer's denial of care. In Colorado, a denial based wholly or partly on medical necessity cannot rest solely on an AI system's output, and a licensed clinician, licensed physician or other regulated professional competent to judge the clinical issues must review and approve it. Georgia lets software take part in utilization review, the check on whether care is necessary, but the software may not issue an adverse determination until a person, with a clinical peer taking part, has reviewed the case. Minnesota bars a review organization from using automated processing alone to deny care. Utah's new law requires whoever decides a denial on medical necessity to use independent medical judgment and not rely solely on any other source's recommendation. That rule does not name AI. All four rules are written about denials, not approvals. Colorado's law also bars coverage of psychotherapy delivered directly by an AI system, in private plans, Medicaid and the state's children's health plan.

Colorado's chatbot law requires any operator of a chatbot open to the public to say that it is AI, keep a protocol for messages about suicide or self-harm and report on it each year to the attorney general. It also bars claims that the chatbot's output comes from, is endorsed by or equals the services of certain licensed professionals. Washington, Oregon and Connecticut set rules for AI companion chatbots, which Oregon defines as systems built to simulate a sustained, humanlike relationship. They must disclose that the user is dealing with AI rather than a person. Oregon and Connecticut require the notice wherever a reasonable person would believe they were talking to a human. All three must point users who voice suicidal thoughts to crisis services. Oregon's definition leaves out software that operates solely to support patient or resident care services in a facility. Connecticut requires operators to keep companions from offering mental health services to users they know or have reason to believe are minors, unless the product is built for that and meets set requirements. In California, a business offering a health chatbot to consumers becomes a provider of health care under the state's medical confidentiality law.

Colorado's attorney general must adopt rules by Jan. 1 clarifying the disclosures owed to a consumer after an adverse decision. In April the Justice Department joined a lawsuit by xAI, an AI company, against Colorado's 2024 AI law, and the state agreed to a court order pausing its enforcement. Colorado replaced that law in May with its new automated decision-making law. A law firm following the case wrote in May that xAI and the Justice Department were expected to challenge the new law as well. How the suit bears on it is not yet clear.

For most physicians, the changes on Jan. 1 will arrive as notices and reviews: a notice to patients in Colorado, a disclosure from an insurer in Utah, a person's review behind a denial in four states. In California they go further, into what software may tell unlicensed staff to do and how practices check their tools for bias.

  • California AB 1979, clinical judgment and health chatbots
    AB 1979 (Bonta), approved by the Governor on Sept. 30, 2026 and chaptered as Chapter 854, Statutes of 2026, takes effect Jan. 1, 2027. It bars a health facility, clinic, physician's office or office of a group practice from using AI to direct, guide, supervise or instruct unlicensed personnel in a clinical function the law requires a licensed person to perform, or to perform such a function independently. The same entities must take reasonable steps so that a licensed health care provider retains the ability to exercise independent professional judgment when care is informed by a clinical decision support system. AI for documentation and communication not involving professional judgment, and use by trainees, are not prohibited. A licensing board may seek an injunction where a violation is the unlicensed practice of a health profession. A business offering a health care chatbot to consumers is deemed a provider of health care under the Confidentiality of Medical Information Act. The Medical Board's analysis covers the July 2 version.
  • California SB 503 and the bills vetoed
    SB 503 (Weber Pierson), approved and chaptered on Sept. 30, 2026 as Chapter 857, Statutes of 2026, takes effect Jan. 1, 2027. It covers developers and deployers of clinical decision support systems, defined as in AB 1979; deployers are health facilities, clinics, physicians' offices and group practice offices that use such a system. Developers and deployers must make reasonable efforts to identify systems that are known or have a reasonably foreseeable risk of biased impacts, meaning diminished access, quality of care or outcomes based on protected characteristics such as race, sex or age. Developers must make reasonable efforts to mitigate that risk and give deployers documentation on training data and demographic representativeness, performance and bias evaluation, data governance and recommended use and monitoring. Deployers must regularly monitor the systems so identified and take reasonable and proportionate steps to mitigate the risk. Gov. Newsom vetoed AB 2575 (Ortega), which would have barred employers from punishing clinicians who override AI recommendations, and SB 903 (Padilla), on AI in psychotherapy.
  • Colorado SB26-189, automated decision-making
    Signed by Gov. Jared Polis on May 14, 2026. Except for provisions listed in Section 5(2), it takes effect January 1, 2027. Under new Section 6-1-1708, Sections 6-1-1701 through 6-1-1706 do not apply to a HIPAA covered entity or its business associates, except for consequential decisions related to employment or an employment opportunity. A covered entity must give patients a general notice of use of advanced technologies. When covered technology determines eligibility for financial assistance, the entity must disclose the decision and the technology's role, the types of information relied on and how to request correction or human review, in advance or within 30 calendar days after an adverse outcome. For a health care provider, the subsection applies only if it operates from a location in Colorado. The attorney general alone enforces the act under the Colorado Consumer Protection Act, after a notice of violation and 60 days to cure unless the violation was knowing or repeated; the cure right sunsets January 1, 2030.
  • Colorado HB26-1139, AI in utilization review
    Session Laws 2026, Chapter 325, approved June 2, 2026, effective January 1, 2027 unless a referendum petition is filed. An AI system used in utilization review must base determinations on the individual's medical or other clinical history, the clinical circumstances presented by the requesting provider and other relevant information in the clinical record, not solely on group data. A carrier's denial of coverage based in whole or in part on medical necessity may not be issued solely on an AI system's output and must be reviewed and approved by a licensed clinician, licensed physician or other regulated professional competent to evaluate the specific clinical issues involved. Carriers, and payers under Medicaid and the Children's Basic Health Plan, may not cover psychotherapy delivered directly by an AI system. The final fiscal note (July 8, 2026) says the rule covers a denial or delay and that the bill bars mental health providers from billing for AI-provided services; the act's text names denials, and its psychotherapy bars bind payers.
  • Insurers' AI in Georgia, Minnesota and Utah
    Georgia SB 444 takes effect Jan. 1, 2027, according to Sheppard Mullin (July 2, 2026), which says it allows AI in utilization review but bars AI from issuing an adverse determination until a natural person qualified as a private review agent or a utilization review entity conducts a utilization review in which a clinical peer participates. Minnesota amended Minn. Stat. 62M.09, subd. 3: from January 1, 2027, a utilization review organization cannot use automated processing alone, and a health care provider must review the case before an adverse determination. Utah SB 319, effective January 1, 2027, amends Section 31A-22-650. An insurer that uses AI in reviewing authorization requests must post a website notice and disclose that use to the department, each in-network provider and each enrollee. The bill's highlighted provisions list a new requirement that a reviewer of an adverse determination exercise independent medical judgment and not rely solely on recommendations from any other source. That subsection does not mention AI.
  • Chatbot laws in Colorado and Washington
    Colorado HB26-1263 was signed May 29, 2026; its requirements take effect January 1, 2027. Operators of conversational AI services accessible to the general public must disclose that the service is artificial intelligence, implement a protocol for prompts about suicidal ideation or self-harm and report on it annually to the attorney general, and may not state that output is provided by, endorsed by or equivalent to services of certain licensed or certified professionals. Washington ESHB 2225 (chapter 168, Laws of 2026; House 69 to 28, Senate 43 to 5, House concurrence 74 to 21) takes effect January 1, 2027. AI companion chatbots must disclose that they are artificially generated and not human, at the start of an interaction and at least every three hours (every hour for known minors or chatbots directed to minors). Operators must keep and post a protocol that refers users expressing suicidal ideation or self-harm to crisis resources.
  • Companion chatbots in Oregon and Connecticut
    Oregon Laws 2026, chapter 85 (SB 1546), approved March 31, 2026, takes effect January 1, 2027. Where a reasonable person would believe they are interacting with a natural person, an operator must give clear and conspicuous notice that the output is artificially generated. Its protocol must use evidence-based methods to detect suicidal ideation and refer users to the national 9-8-8 suicide and crisis lifeline. The definition excludes software that operates solely for purposes such as assisting or supporting patient or resident care services in a facility. Connecticut's SB 5 became Public Act 26-15, signed May 27, 2026; its sections on AI companions (Sections 4 to 6) take effect January 1, 2027. Operators must refer users at risk of suicide or self-harm to resources such as 988, give notice where a companion would cause a reasonable individual to believe they are interacting with a human, and keep companions from offering mental health services to users they know or have reason to believe are minors, unless designed to deliver them and meeting specified requirements.
  • The federal order and the Colorado suit
    Executive Order 14365, signed Dec. 11, 2025 and published at 90 FR 58499 on Dec. 16, 2025, directed the Attorney General to establish within 30 days an AI Litigation Task Force whose sole responsibility is to challenge state AI laws inconsistent with the order's policy. xAI sued on April 9, 2026 over SB24-205, Colorado's 2024 AI law, in the U.S. District Court for the District of Colorado; the Justice Department moved to intervene on April 24, and the court granted the intervention. Under a stipulated order of April 27, 2026, Colorado's attorney general agreed not to begin enforcement or investigations until 14 days after the court rules on xAI's motion for a preliminary injunction, which xAI would not file until 28 days after final rules under SB24-205 or replacement legislation such as SB26-189. Colorado adopted SB26-189 on May 14, 2026, replacing the 2024 law. The law firm McDermott wrote on May 27, 2026 that xAI and the Justice Department are expected to challenge SB26-189 as well.

Latest news

  1. Medicare Advantage plans would have to make in-network clinicians file prior authorizations through an electronic interface, CMS proposes
  2. Medicare's first mandatory Advantage model would swap star ratings for local 'Diamond Ratings' in randomly chosen states from 2029
  3. A California health system says an employee uploaded emergency patients' diagnoses and record numbers to ChatGPT

See all the latest news