Physician in the Loop

CMS chat room gave tech firms access to officials, report says; device makers seek wider RAPID path; FDA grants pacemaker identifier; GPT-5 misreads images

Show Notes

CMS chat room gave tech firms access to officials, report says; device makers seek wider RAPID path; FDA grants pacemaker identifier; GPT-5 misreads images

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

Technology and AI companies lobbied federal health officials for about a year in a 1,700-member Slack workspace run by the Centers for Medicare & Medicaid Services, KFF Health News reported Oct. 9.

Only a handful of the members were patient advocates, physicians or hospital representatives, according to the report, which drew on thousands of messages, meeting transcripts and recordings. Vendors in the workspace pushed for broader access to patients' records for vetted apps, and CMS officials indicated that apps the agency promotes would be prioritized for a new program paying for AI chatbot advice and wearable tracking, the outlet reported. CMS did not answer questions about whether the group complies with federal advisory committee law.

The weekend was quiet. Device makers pressed CMS to widen its proposed RAPID coverage pathway before comments close Oct. 13, the Food and Drug Administration granted a De Novo to a handheld identifier for implanted cardiac devices, and a study of GPT-5 found image misreadings and unreliable web search in real-world diagnosis.

In this episode

What is Physician in the Loop?

How technological innovation is changing medicine for patients and the people who practice on the front lines.

Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Monday, October 12th, 2026.

Voice 1: For about a year, a Slack workspace run by CMS gave AI and tech companies direct access to Medicare and FDA officials. It had about 1,700 members, and only a handful were patient advocates, physicians or hospital representatives.

Voice 2: Who was actually in that room?

Voice 1: Oura, Palantir and the venture firm 8VC were among the members. The workspace opened in August 2025, run by Amy Gleason, now chief product officer of CMS's Office of Health Technology and Products. Through it, the FDA invited Microsoft, Anthropic, OpenAI, Apple and Google to a February listening session on conversational AI products that was not on its public calendar.

Voice 2: What were companies actually pushing for in there?

Voice 1: Broader access to patients' records for vetted apps. CMS officials also indicated that apps listed in its Medicare App Library, which promotes about two dozen commercial apps, would be prioritized for a new program paying companies for AI chatbot advice or wearable tracking.

Voice 2: Is a group like that even allowed to operate this way?

Voice 1: That's disputed. Joseph Daval, a former FDA lawyer now at Harvard Medical School, said the group resembles a federal advisory committee, which has to be open to the public. Gleason called it an open, voluntary technical collaboration, and the workspace's own code of conduct says it isn't an advisory committee. CMS did not answer whether it complies with federal advisory committee law.

Voice 2: Did any physician group weigh in?

Voice 1: The AMA's chief executive, John Whyte, told the outlet some AI tools aren't ready for primetime, in his words.

Voice 2: A follow-up on the RAPID pathway story: AdvaMed's letter has more detail now.

Voice 1: It does. CMS expects about 40 devices would qualify for RAPID right away, with about 20 more later. Under the proposal, Class II devices would have to join the FDA's Total Product Life Cycle Advisory Program to be eligible.

Voice 2: And diagnostics, since AdvaMed asked for those to be included?

Voice 1: CMS does not plan to cover in vitro diagnostics through this pathway. Comments close tomorrow, October 13th.

Voice 2: A different kind of FDA action this week, for cardiac emergencies. The agency granted a De Novo authorization to a device called SmartRead.

Voice 1: What does it actually do?

Voice 2: It's a handheld device from BridgeSource Medical, in Austin, Texas. It mimics the opening signals of each major manufacturer's programming wand, including Abbott, Biotronik, Boston Scientific, Medtronic and MicroPort, and waits for an implanted pacemaker or defibrillator to answer. In about 20 seconds, it tells emergency department clinicians which programmer that patient's device needs.

Voice 1: Does it run on AI?

Voice 2: The report doesn't say. It also cites no clinical performance data. Development was supported by small business grants from the National Institute on Aging, including $3 million in commercialization funding, according to a co-founder.

Voice 1: A study out this week put GPT-5 through 161 real-world diagnostic cases, and the images were its weak point.

Voice 2: So the reasoning held up but the pictures didn't?

Voice 1: That's about right. Researchers from the University of Pennsylvania, Stanford, Mayo Clinic, the National Library of Medicine and other institutions tested 10 GPT-5-based configurations, including agentic and web-search setups, on cases that combined text and images. The models used patient histories well and gave clear stepwise reasoning with plausible differentials, but they misinterpreted images, and inconsistent web search sometimes lowered performance.

Voice 2: Did the authors say what actually helps, then?

Voice 1: They wrote that reliable multimodal diagnosis depends more on transparent reasoning and high-quality inputs than on aggressive retrieval or complex agentic workflows. The paper is an accepted version that may still change before final publication.

Voice 2: Before we go, something to watch.

Voice 1: The CPT Editorial Panel's latest summary lists a Category I code for AI arrhythmia analysis, effective January 2028, plus three AI Category III codes effective January 1st, 2027.

Voice 2: And on the coverage side?

Voice 1: Palmetto GBA and Noridian have proposed Medicare noncoverage of AI brain MRI quantification. Comments on that are due November 14th.

Voice 2: There's also a sooner deadline: comments on the FDA's generative-AI device discussion paper are due October 19th.

Voice 1: That's the docket covering how the agency should think about generative AI in devices generally, not tied to any one product.

Voice 2: Between that paper, the RAPID comments closing tomorrow, and the CPT codes still a year or more out, physicians are being asked to weigh in on AI payment and coverage from several directions at once, with none of it settled yet.

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.