Wednesday, September 9, 2026

1. Signal of the Day · Regulation & Policy

   New bipartisan bill would bar AI from issuing medical-necessity denials — but not in Medicare Advantage or Medicaid

   H.R. 10210, the Doctors Not AI Act of 2026, introduced September 1 by Rep. Greg Landsman with three same-day cosponsors, would require that any adverse benefit determination involving clinical judgment not be issued or dictated by an AI system, be made by a licensed professional with training in the service at issue, and not treat AI output as presumptively valid. It adds disclosure duties and an administrative-record duty covering the system used, its role, and any outputs or scores it generated, discoverable by the enrollee on request, and makes AI use in utilization review a treatment limitation for parity purposes. Its definition of a determination involving clinical judgment expressly includes level of care and health care setting — the post-acute fact pattern. But it amends title XXVII of the Public Health Service Act, ERISA and the Internal Revenue Code, and the full bill text contains zero occurrences of Medicare and zero of Medicaid, so the payers behind most skilled-nursing denials fall outside it as drafted.

   HighBy govinfoRead at source ↗

2. AI Tech & Frameworks Surfaced late · study published June 11

   Largest study of ambient AI scribes finds modest documentation savings and no measured productivity gain

   Across 198,178 emergency department encounters at four hospitals in a large integrated system, ambient AI scribes were associated with a 1.6-minute reduction in adjusted median attending documentation time per note, compared with 3.3 minutes for human scribes, and total work relative value units per shift hour did not differ among groups. Two counterweights keep this from reading as evidence that scribes do not work: a July 2026 cohort study (PMID 42391625) found a 72.6-second per-encounter reduction, roughly 24 minutes across a 20-encounter shift, and a September 2026 scoping review (PMID 42585863) found only six qualifying emergency department studies with physician samples of six to 24, one reporting longer charting time with AI than with human scribes. As ambient AI is sold into structured clinical documentation, the practical lesson is to build the business case on measured documentation minutes rather than on throughput, because throughput is the claim the evidence does not currently support.

   HighBy Annals of Emergency MedicineRead at source ↗

3. Security & AI Risk

   Two ransomware groups claim the same nationwide home-health provider, one leaking franchise governance records

   Interim HealthCare, operating in 40 states, was listed by the Genesis group on August 10 claiming 1TB including medical records, patient lists and clinical data, and by Anubis on August 21 claiming 530GB including franchisee financial information, internal and external audit records and operational memoranda, with samples published indicating no ransom was paid. Interim HealthCare of Oklahoma City reported a network-server hacking incident to the HHS Office for Civil Rights on July 31, 2026 using a placeholder estimate of 500 individuals. There is no AI dimension to this incident. For franchised or multi-entity post-acute organizations the second listing is the more unusual problem: the exfiltrated set is corporate-governance material about the franchise system, which raises different disclosure and contractual questions than a patient-data breach.

   MedBy Steve Alder · HIPAA JournalRead at source ↗

4. SNF / Post-Acute

   Behavioral-health acuity and antipsychotic scrutiny collide with survey-eligibility criteria in nursing homes

   Clinical leaders describe a coming wave of behavioral-health acuity in nursing homes, arriving alongside a CMS schizophrenia-misdiagnosis audit and an Office of Inspector General work-plan item on antipsychotic use, with operators warned that a misdiagnosis finding can suppress quality measures in totality. The item carries no AI content. It is worth reading next to survey policy because post-admission schizophrenia coding is one of the enumerated criteria that disqualifies a facility from the new risk-based survey pathway, excluding 2,867 facilities or 15.22 percent as of June 2026 data, per Appendix D of QSO-26-14-NH. Coding-integrity exposure and survey-eligibility exposure are increasingly the same exposure.

   LowBy Skilled Nursing NewsRead at source ↗

5. SNF / Post-Acute

   MatrixCare's new CEO puts voice-enabled MDS documentation and AI trend detection at the top of the roadmap

   Incoming CEO Jonathan Lujan said the post-acute EHR and revenue-cycle vendor will prioritize practical AI and automation for clinical documentation, revenue cycle and workflows, naming two skilled-nursing specifics: streamlining MDS documentation through voice-enabled tools, and using AI to identify clinical trends such as increased falls for earlier intervention. The governance distinction worth holding is that the MDS is not a narrative note but a payment- and survey-consequential structured assessment, now subject to an automated accuracy audit. Moving ambient or voice AI into that tier shifts it from low-stakes documentation into the data that determines reimbursement and survey eligibility, which makes attestation of the coded output the control that matters.

   MedBy Zahida Siddiqi · Skilled Nursing NewsRead at source ↗

6. Regulation & Policy

   CMS renames and expands the hospice assessment system of records, adding real-time data collection

   Published September 9, the notice modifies System No. 09-70-0548, renaming it from Hospice Item Set to Hospice Outcomes and Patient Evaluation, with the stated reason being inclusion of real-time data collection at the time of patient assessments rather than only at admission and discharge. Categories of records now add ethnicity, preferred language and the Medicare Beneficiary Identifier, and routine use 4 is rewritten to authorize disclosures supporting research and payment-initiative analysis. The notice contains no AI content and should not be read as an AI item. It matters as pattern: more assessment data, collected more often, flowing to CMS under disclosure authority written broadly enough to support downstream analytics — the same dynamic that made assessment-data accuracy consequential in skilled nursing.

   MedBy Federal RegisterRead at source ↗

Get the Daily Dispatch by email · free, about 8 a.m. Eastern · RSS