Doctors Not AI Act Would Bar AI From Clinical Denials in Employer and Individual Health Plans
On September 1, 2026, Rep. Greg Landsman (D-OH-1) introduced H.R. 10210, the Doctors Not AI Act of 2026, with Reps. Earl L. “Buddy” Carter (R-GA-1), Kim Schrier (D-WA-8), and Tom Barrett (R-MI-7) as original cosponsors. The bill would amend the Public Health Service Act, ERISA, and the Internal Revenue Code so that a group health plan or an issuer of group or individual coverage could not let an AI system issue or decide an adverse benefit determination involving clinical judgment.
The duty appears three times, in new PHSA section 2719B, a new paragraph (3) in ERISA section 503, and new Internal Revenue Code section 9815A. Each requires the plan, and under the PHSA the issuer as well, to “ensure” that any such determination, “whether initially or upon appeal,” is “not issued by an artificial intelligence system and is not dictated or determined by the output of such system.” It must be made “only by a licensed health care professional” acting within the scope of that license, with “training and experience” in the item or service at issue, who “shall not treat any output of an artificial intelligence system as presumptively valid.”
An “adverse benefit determination” includes “an initial determination and a determination on internal appeal.” One “involving clinical judgment” is any determination based “in whole or in part” on medical necessity, appropriateness, experimental or investigational exclusions, level of care, setting, effectiveness, clinical guidelines, utilization review criteria, or other standards that turn on the enrollee’s condition or treatment needs. The AI definition reaches any algorithm, predictive model, or automated decision-making software “used to materially influence, automate, or issue determinations regarding coverage.”
The section then adds a notice duty and a record duty. Where an AI system “was used in connection with the determination,” the denial notice that section 2719 requires must say so, describe the system’s “role in the review process,” and give “the name, professional license, and credentials” of the professional who decided. The administrative record must hold documentation describing the system, its role, and “any outputs, scores, recommendations, or determinations generated by such system.” Those materials “shall be considered part of the administrative record and made available to the enrollee upon request.”
What the text changes is records a plan already keeps. Denial notices would need fields for AI use and a named clinician, and claim files would have to retain model scores and recommendations an enrollee could ask to see.
Each of the three parity sections gains one sentence: “The use of an artificial intelligence system in connection with utilization review shall constitute a treatment limitation for purposes of this section.” Each also gains a clause (vi) in subsection (a)(8)(A), the list of what a plan’s comparative analysis of nonquantitative treatment limitations must contain. The new clause asks whether an AI system “is used in, or materially influences, the design, development, application, or administration of such limitation.” If so, the analysis must give the Secretary enough information to evaluate its “function, operation, and effects” on mental health and substance use disorder benefits against medical and surgical benefits, “both as written and in operation.” On our reading, those two amendments together would bring any AI system a plan subject to parity uses in utilization review inside the parity comparative analysis.
In July the Senate declined to take up a Congressional Review Act resolution aimed at CMS’s WISeR model in traditional Medicare, where CMS says final denials are made by licensed clinicians, not machines. Senators have also pressed three Medicare Advantage insurers for records on how algorithms figure in their coverage denials. H.R. 10210 names neither program. State laws such as Minnesota’s sit in state insurance codes; this bill writes the duty into ERISA itself. AI in health insurance maps the wider federal and state picture.
Schrier said in her release: “Medical decisions should be made by health professionals, and we need oversight and transparency now to keep Americans safe and healthy.”
On introduction the bill was referred to Energy and Commerce, with additional referrals to Ways and Means and Education and Workforce. The status record, last updated September 3, shows no further committee action and no related bills. Whether a Senate companion or hearing exists, and where AHIP or other trade groups stand, remains unverified.
Section 5 applies the amendments “to plan years beginning on or after January 1 of the first calendar year beginning not less than 12 months after the date of the enactment of this Act.” Everything in that sentence is measured from an enactment date, and the record has none.
Official document
govinfo.gov →The instrument itself, issued by a government, court, legislature, or standard-setting body.