Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Tuesday, November 17th, 2026.
Voice 1: Medicare wants every clinician who bills a Medicare Advantage plan to also hold active enrollment in traditional Medicare, a proposed fraud rule CMS says would save nearly $90 billion.
Voice 2: Nearly $90 billion riding on an enrollment requirement. What does getting enrolled in traditional Medicare actually put someone through?
Voice 1: Identity verification, a licensure review, site visits, and fingerprint background checks. CMS is calling this the CRUSH rule, short for Comprehensive Regulations to Uncover Suspicious Healthcare, issued November 16th. The Advantage enrollment requirement itself wouldn't start until the plan year beginning two years after a final rule is published.
Voice 2: Two years out. Is there anything in here that actually lands on providers sooner than that?
Voice 1: Yes. Every Medicare provider, not just those billing Advantage plans, would get only 30 days to report a change to their enrollment information, down from as many as 90. CMS could also deny or revoke enrollment outright if an owner, manager, officer or director's identity can't be confirmed.
Voice 2: What's CMS's own side of the ledger look like, against that nearly $90 billion in savings?
Voice 1: About $380 million in costs from 2027 through 2036, by the agency's own estimate. And CMS said flatly it intends to put out a program integrity rule like this one every single year from now on.
Voice 2: A study out of Germany this week found something uncomfortable about leaning on AI. As researchers made an AI assistant fail on purpose, physicians' own accuracy fell right along with it, and almost nobody noticed without being asked.
Voice 1: Almost nobody noticed. How many people are we talking about, and how bad did the AI's reliability actually get?
Voice 2: Forty-eight participants, including 18 final-year medical students, classified 40 heart sounds while researchers dropped the tool's reliability in stages, from 90% down to 60%, then 50%, then 0%. Accuracy with the tool's help fell from about 81% down to about 58%.
Voice 1: Did it help at all when researchers made people commit to their own answer before seeing what the tool said?
Voice 2: No, it didn't improve performance. And their confidence in their own judgment actually fell faster than their trust in the tool kept falling.
Voice 1: So out of everyone in the study, how many actually caught the tool breaking down without being prompted to look for it?
Voice 2: About one in eight. The researchers, at Philipps-Universität Marburg, put it plainly: human oversight alone “may be insufficient to reliably detect system malfunctions.”
Voice 1: A trial out this week pushed semaglutide down into much younger patients. In children six to under twelve with obesity, once-weekly semaglutide cut body mass index by about 14% over 68 weeks, while kids on placebo saw their BMI rise about 3%.
Voice 2: That's a wide gap for kids that age. How many children were actually in this trial?
Voice 1: The STEP Young trial randomly assigned 167 children, two to one, with lifestyle counseling in both groups. 38% of the children on semaglutide ended up below the 95th percentile threshold for obesity, against just 2% on placebo.
Voice 2: Did that gap hold up over time, or close later on?
Voice 1: It widened. By 104 weeks, the difference between the groups reached nearly 18 percentage points. Gastrointestinal side effects were the most common problem, more frequent on semaglutide, and led one child to stop treatment.
Voice 2: Semaglutide's already approved for obesity in adolescents and adults. Worth saying Novo Nordisk, which makes the drug, funded this trial.
Voice 1: Funded or not, a gap that keeps widening two years in is the kind of result that gets noticed.
Voice 2: Measles turned up in a new place to worry about this week. CDC counted 112 confirmed measles cases during pregnancy, and 13 cases of congenital measles, from January of last year through late September.
Voice 1: Congenital measles in newborns. What complications actually came with that?
Voice 2: Maternal pneumonia, preterm delivery, and neonatal complications, according to the CDC's preliminary analysis.
Voice 1: Were most of these patients at least partly protected going in?
Voice 2: No. 98% of the pregnant patients hadn't gotten an MMR vaccine before pregnancy, or their vaccination status was unknown.
Voice 1: Is there anything else driving these cases, or does it really come down to that one gap?
Voice 2: The sources don't point to anything else. It comes down to that same vaccination gap, case after case.
Voice 1: Before we go, something to watch. The Senate's health committee votes November 18th on Heidi Overton, the nominee for FDA commissioner.
Voice 2: What actually happens if that vote goes her way?
Voice 1: Approval sends her nomination on to the full Senate, according to STAT.
Voice 2: One committee vote away from knowing who's actually running the agency behind rules like today's enrollment proposal.
Voice 1: Comments are also due November 24th on FDA's draft guidance for robotically assisted surgical devices.
Voice 2: What's that guidance actually asking for?
Voice 1: A risk assessment of AI features built into those devices, under docket FDA-2026-N-9505. The draft leaves out remote telesurgery and autonomous surgical robots entirely.
Voice 2: A guidance that draws its own line around what counts as robotic surgery, and leaves the fully autonomous version sitting outside it for now.
Voice 1: Sources and corrections are at physicianintheloop.org, where you can sign up for the free email briefing. This briefing was read by AI voices.