Kennedy calls AI better informed than any doctor; McKinsey sees AI in up to 22% of outpatient claims; Oura postpones IPO; Penn Medicine adds OpenEvidence
Listen to this post
all episodesTHE SHORT READ
Health and Human Services Secretary Robert F. Kennedy Jr. told the Make America Healthy Again Summit in Washington on Sept. 29 that artificial intelligence can give Americans a second opinion better informed than any doctor.
Kennedy also said OpenAI CEO Sam Altman had told him it would be "malpractice" to diagnose without consulting AI, Forbes reported. A day earlier, McKinsey & Co. estimated that AI could perform the work behind 16% to 22% of U.S. outpatient claims, from low-complexity office visits to imaging interpretation. Smart ring maker Oura postponed its IPO on the eve of trading, and Penn Medicine opened OpenEvidence to 10,000 clinicians.
In California, Gov. Gavin Newsom's deadline to act on four health-AI bills, including limits on AI marketed as therapy, falls today. A University of Florida study found that five AI evidence-search tools each missed large shares of the published evidence on a clinical question.
THE DETAILS
- Regulation Sept 29, 2026Health and Human Services Secretary Robert F. Kennedy Jr. said at the Make America Healthy Again Summit in Washington on Sept. 29 that artificial intelligence can give patients "a second opinion that is much better informed than any doctor in the country" and can "free us from medical tyranny," Forbes reported. Kennedy said OpenAI CEO Sam Altman had told him it would now be "malpractice" to diagnose or prescribe without consulting the technology, according to Forbes and The Hill. Vice President JD Vance said experts no longer hold the same "monopoly on knowledge" and that a company whose product harms people "should suffer consequences for it," the Washington Examiner and Forbes reported. Centers for Medicare & Medicaid Services Administrator Dr. Mehmet Oz said the health system is not using AI enough and acknowledged its risks, according to The Hill, which reported that nine publicly traded companies, AI firms among them, sponsored the event. No new policy was announced.
- Workforce Sept 28, 2026A McKinsey & Co. report published Sept. 28 estimated that AI can perform the work behind 2 billion to 3 billion U.S. outpatient claims a year, or 13% to 19% of outpatient spending, based on 2024 commercial, Medicare and Medicaid claims. Low-complexity evaluation and management visits account for 11% to 15% of claims, and diagnostic and imaging interpretation, identified by the professional-component modifier 26, for 5% to 7%, or 700 million to 900 million claims, the authors wrote. The analysis counts only "discursive care," the reasoning and dialogue work of intake, triage, diagnosis, referral and follow-up that needs no hands-on exam, and says AI can perform a substantial share of core tasks for 11 million of 19 million health care workers. The authors wrote that fee-for-service organizations "risk losing low-acuity-visit volume" but said rising demand means the workforce is not projected to shrink, Becker's reported. The figures estimate what AI could do, not care it now delivers.Sources: McKinsey & Co., Becker's Hospital Review
- Money Sept 29, 2026Smart ring maker Oura said Sept. 29 that it postponed its initial public offering despite strong demand, citing uncertainty in the IPO market, and set no new date. The company and existing stockholders had planned to raise about $2.2 billion selling shares at $40 to $44, Fierce Healthcare reported, and HIStalk reported that trading had been set to begin the next day. Oura said it is profitable, has 5.7 million paid members and expects revenue to grow 90% in the fiscal year that ends Sept. 30; the figures are the company's own. "We have the luxury of choosing our moment," CEO Tom Hale said. The company's registration statement with the Securities and Exchange Commission has not been declared effective.
- Deployment Sept 28, 2026Penn Medicine is integrating OpenEvidence's clinical decision-support tool across its seven-hospital system for 10,000 clinicians, Fierce Healthcare reported Sept. 28, and will build customized tools with clinicians of the Botswana-UPenn Partnership, a 25-year academic collaboration. "Effective clinical AI will be shaped by local expertise," said Dr. Farouk Dako of Penn Medicine. The announcement came with Anthropic's commitment of credits, engineers and funding for a free version of the tool in about 100 countries, first reported Sept. 24. OpenEvidence says 1.12 million verified U.S. clinicians use it and that it handled 42 million consultations in August; the figures are the company's own.Source: Fierce Healthcare
- Evidence Sept 29, 2026University of Florida researchers tested five AI evidence-search tools, Consensus, Ai2 Paper Finder, ChatGPT, Gemini and Claude, against a non-public gold-standard set of studies using 15 ways of phrasing the same question, according to a paper published Sept. 29 in npj Digital Medicine. Recall for a single phrasing ranged from 7.2% to 42.2%, and pooled across all phrasings from 45.8% to 72.3% by platform; a single query retrieved none of the target evidence 47% to 80% of the time. No platform ever retrieved 12% of the evidence, and conference proceedings went unretrieved far more often than journal articles, 38.9% against 4.6%, the authors reported. OpenEvidence was not among the tools tested. The authors recommend evaluating such tools within a clinical domain before relying on them.Source: npj Digital Medicine
- Evidence Sept 29, 2026Anesthesiologists led by Tao Yan of China's National Cancer Center built a model from CT-derived lung volumes measured at two time points plus clinical features to predict intractable hypoxemia during one-lung ventilation in 2,040 patients who had repeat lung resections at seven Chinese hospitals from 2013 through 2023, according to a paper published Sept. 29 in npj Digital Medicine. Hypoxemia occurred in 15.4% of patients. The model reached an area under the curve of 0.790 in training and 0.750, 0.765 and 0.782 in three validation cohorts, while models built on pulmonary function tests alone scored below 0.6, the authors reported. They said prospective validation is needed before clinical use.Source: npj Digital Medicine
- Deployment Sept 28, 2026Cooper, the Camden, New Jersey, health system, began piloting Microsoft's Dragon Copilot ambient documentation tool for nurses on one unit at its university hospital, Becker's reported Sept. 28, citing the system's release. Nurses record conversations and spoken observations on mobile devices after patients consent, and the tool drafts structured record entries that nurses review, edit and approve. The system already holds more than 1,300 Dragon Copilot licenses for physicians and advanced practice providers and will measure documentation time, accuracy and nurse satisfaction before deciding on expansion; no results have been reported.Source: Becker's Hospital Review
- Regulation Sept 29, 2026Sept. 30 is the last day for Gov. Gavin Newsom to act on bills passed at the end of California's 2026 session, and LegiScan's records as of Sept. 29 showed no action on four health-AI bills presented to him: AB 1979 and AB 2575, both titled Health Care Services: Artificial Intelligence; SB 503, on AI in health care services; and SB 903, on AI and mental health professionals. SB 903 would bar advertising AI chatbots as therapy and set guardrails for AI in mental health treatment; it passed the Senate unanimously and the Assembly 71-4, and thousands of licensed mental health professionals signed a letter urging a signature, Sen. Steve Padilla's office said. The bills bind no one until they become law.
- Regulation Sept 29, 2026The Senate Health, Education, Labor and Pensions Committee scheduled votes Wednesday on three Health and Human Services nominees but not on Dr. Heidi Overton, the nominee to lead the Food and Drug Administration, and her confirmation vote has been put off until after the midterm elections, STAT reported Sept. 29. The committee heard her nomination Sept. 24.Source: STAT