ยท Pinned
Physician in the Loop launches with a September 2026 baseline on AI in physician work
THE SHORT READ
Physician in the Loop launched this week with a baseline of where artificial intelligence stood in physician work as of September 2026, drawn from the regulatory record and published evidence. The longer analysis is in the "Where things stand" section of the site.
Federal officials have discussed a new Medicare payment category for AI clinical software and rates for company-operated AI physicians at 60% to 80% of the human fee, The New York Times reported Sept. 14; nothing has been proposed. In trials published this year in Nature and Science, models matched or outperformed physicians on text-based tasks; the Nature paper called its system "not ready for real-world translation." No published trial has shown an AI model improving patient outcomes on its own. Twenty-two states have enacted 33 health-AI laws this year, according to Manatt Health's tracker, and no state has authorized an AI to practice medicine.
THE DETAILS
- Regulation Sept 14, 2026The New York Times reported Sept. 14 that federal officials have discussed a new Medicare payment category for AI clinical software, with rates for AI physicians operated by technology companies at 60% to 80% of the human fee. Nothing has been proposed. Dr. John Whyte, chief executive of the American Medical Association (AMA), said autonomous models are not ready.
- Regulation Aug 18, 2026The Food and Drug Administration (FDA) issued a discussion paper Aug. 18 proposing a two-axis risk framework and "competency-based" premarket evaluation for generative-AI devices, with postmarket monitoring. The agency's digital health chief said formal guidance is coming. The docket is open until Oct. 19.Source: FDA
- Regulation Sept 15, 2026Medicare's ACCESS model, which launched in July and pays for technology-enabled care of hypertension, diabetes, musculoskeletal pain, depression and anxiety, now has about 160 participating organizations and adds heart failure, COPD, substance use and tobacco next spring, the Centers for Medicare & Medicaid Services (CMS) said. Paired with the FDA's TEMPO pilot, tools that have never been cleared, including a voice-AI therapy program, are treating beneficiaries under enforcement discretion in exchange for data. The CMS administrator described the program as access to "AI-enabled technologies."
- Deployment Sept 9, 2026The Advanced Research Projects Agency for Health (ARPA-H) is paying $62.7 million to Kaiser, Duke, Stanford, Tempus, UpDoc and Atman to build agents that assess symptoms, titrate guideline-directed therapy and order labs inside a physician-prescribed plan, with mandatory escalation and FDA authorization packages due within two years, STAT and Fierce Healthcare reported. Kaiser will run a trial of about 2,500 patients. A health system chief medical information officer described the approach as the shift from human-in-the-loop to exception-based oversight.
- Regulation Sept 15, 2026About 1,000 pages of records obtained by the Electronic Frontier Foundation (EFF) show that one vendor in Medicare's WISeR prior-authorization pilot told CMS before launch it would auto-affirm requests because it was not ready, two vendors denied more than 20,000 requests in three months, vendors are paid per denial and one request sat for 83 days. A Senate vote to end the program failed 46-50 in July.
- Deployment Jul 30, 2026Akido Labs describes itself as the first AI-native health system, with 1 million patients, 100 clinics, medical assistants running AI-guided visits with a physician approving remotely and, by its own figure, four to five times the patients per physician. No outcomes trial has been published. Hims & Hers said AI answers 80% of the questions from weight-loss patients on its new care platform. Doctronic's Utah refill sandbox remains at 100% physician sign-off after seven months.
- Evidence Jun 17, 2026Google's AMIE trial in Nature was a randomized, blinded virtual OSCE with 100 multi-visit cases and 21 primary care physicians; the model's management plans were rated appropriate 95% to 98% of the time versus 72% to 81% for the physicians. The same paper describes the system as "not ready for real-world translation," citing a text-only format with patient actors and no pharmacist or order entry. In April, Science published a Harvard study in which an OpenAI model matched attending physicians on emergency department triage and admission decisions using real records. Stanford researchers have repeatedly found that a physician using the model does no better than the model alone, and that the gap closes only when the physician commits to an assessment first and then compares.
- Deployment Aug 20, 2026At its August user group meeting, Epic shipped Ergo Visit, live at Ochsner since Aug. 3, in which the assistant drafts the note and the orders for the physician to review; an Agent Factory for building no-code agents, widely available in 2027; and a generative model trained on 310 million patients' records to predict readmission and stroke risk. Oracle extended its clinical agent to nurses this week, and OpenAI's ChatGPT for Healthcare now reads Epic charts at UCSF, Healthcare Dive and Fierce Healthcare reported.
- Workforce Aug 28, 2026Doximity's August compensation report put overall physician pay up 2%, with family medicine at $325,040 and emergency medicine at $423,723; 85% of physicians reported shortage effects and 46% were considering early retirement. It also found 20% already face higher productivity targets because of AI and 42% expect them, while 24% expect AI to raise their pay. The Medical Group Management Association (MGMA) found compensation rising in all 23 specialties it tracks while work RVUs fell in 16 of them.
- Regulation Aug 12, 2026Manatt Health's tracker counts 33 health-AI laws in 22 states through July. The laws typically require AI disclosure to patients, licensed accountability for clinical use, human review of payer denials and a bar on chatbots posing as therapists. Colorado's HB 1177 took effect Aug. 12 and requires licensed oversight of any clinical AI. California's AB 1979, which would bar AI from performing licensed work or directing unlicensed staff, sits on the governor's desk until Sept. 30. No state has authorized an AI to practice medicine, and of 20 liability bills introduced, none passed.Source: Manatt Health
- Money Sept 15, 2026OpenEvidence, free to clinicians and ad-supported, is valued at $12 billion; in July it considered, then shelved, a round at a $20 billion valuation. Function Health, which sells a $365-a-year lab bundle, raised another $450 million in July. Oura filed to go public in September with $1.4 billion in trailing revenue. Doximity's shares rose by roughly two-thirds after it said its AI scribe users grew tenfold. Waystar, which sells billing software, hired bankers to explore a sale after investors concluded AI would displace its business.
- Incident Sept 3, 2026Winters v. OpenAI, filed July 23, alleges the chatbot told a man with unstable blood pressure to stay in his recliner and he developed a pulmonary embolism; the complaint pleads unlicensed practice of medicine, CBS News reported. No malpractice suit over an AI scribe has been filed, Healthcare Dive reported, but malpractice carriers are already describing the failure modes: hallucinated template text, omitted nonverbal findings and the decay of review rigor once the clinician trusts the draft.