PharmaOpinion
AMA CEO John Whyte Warns AI Oversight Can Become “an Illusion”
The federal Clinical Super Intelligence Roadmap calls for AI to start with lower-risk tasks and take on more responsibility as evidence from real-world care supports it.

The brief
- CMS wants clinical AI to start with lower-risk tasks and earn more autonomy through evidence from real-world care.
- AMA CEO John Whyte says care teams should decide before deployment who gets alerted, who can override AI and who follows up with the patient.
- Medicare’s ACCESS model ties full payment for hypertension care to measurable improvement or control of blood pressure.
A new reality is emerging in healthcare. If AI can change your medication between visits, who decides when your doctor needs to step in?
The brief
- CMS wants clinical AI to start with lower-risk tasks and earn more autonomy through evidence from real-world care.
- AMA CEO John Whyte says care teams should decide before deployment who gets alerted, who can override AI and who follows up with the patient.
- Medicare’s ACCESS model ties full payment for hypertension care to measurable improvement or control of blood pressure.
American Medical Association (AMA) chief John Whyte put it best when he wrote on LinkedIn, “Accountability must follow control.” He used high blood pressure as an example. In one scenario, AI flags a rise in a patient’s readings and recommends a dose change for the doctor to review. In the other, AI changes the dose and decides whether to alert the doctor.
Whyte was responding to the Clinical Super Intelligence Roadmap, which the Centers for Medicare & Medicaid Services (CMS) released a few days ago.
The roadmap outlines how AI-led primary care could take part in federal payment experiments. Whyte argued that before software gets more authority over treatment, it needs evidence that it works in everyday care, and the care team needs to know who gets notified, who can intervene, and who follows up with the patient.
Who Gets the Call?
CMS outlined a phased approach that starts with lower-risk uses and expands autonomy as evidence from clinical practice supports it. Under the roadmap, AI-led services participating in future payment models would need partnerships with clinicians and facilities so patients could reach a human clinician when needed.
RxRoundtable Founder Harry Travis argued that physicians could supervise AI without reviewing each decision, allowing primary care teams to reach more patients. Mary Gress, who works on AI governance in health care, asked who on the care team would be notified when the software changed a dose and who would follow up.
“Who is notified, who has authority to intervene, and who is responsible for follow-up?” Whyte opined. He said those responsibilities should be established before deployment. He also questioned how much oversight physicians could provide if they were expected to verify hundreds of AI-generated decisions.
The AMA’s framework for clinical AI liability ties responsibility to what each party can control and what information it can access. A developer can fix how software interprets a blood pressure reading. The organization deploying it can decide where the alert goes. The physician’s ability to intervene depends on receiving it in time and seeing the information behind the recommendation.
Medicare Is Testing Payment for Blood Pressure Outcomes
A separate Medicare experiment already ties payment to how well patients’ chronic conditions are managed. Under the Advancing Chronic Care with Effective, Scalable Solutions (ACCESS) Model, participating organizations receive recurring payments, with full payment for hypertension tied to measurable improvement or control of blood pressure.
Each participating organization must designate a physician medical director responsible for care quality and compliance. Medicare also pays eligible referring clinicians to review updates and coordinate care, giving that work a payment of its own.
What Oversight Looks Like Between Visits
Lateef Cannon, a family physician and IT director at The Family Health Centers in Asheville, North Carolina, proposed a third scenario in response to Whyte. AI could help patients navigate care between visits, while nurses and medical assistants working within the physician’s care plan could decide which patients need the doctor’s attention.
“Our nurses and MAs know which patients worry, which ones get sick fast and who has support at home,” said Cannon. In his proposed workflow, that knowledge would inform which patients staff escalate and what the physician sees when they take the call.
For a practice testing Cannon’s approach, that means tracking which cases staff catch that the software misses and how long it takes an alert to reach someone who can act. Reviewing those alerts takes staff time, which is why Cannon would start by using AI to free up some of it. “AI’s first job is taking clerical work off everyone’s plate,” he said.
Join the discussion
0 answers
People who run these systems answer here. Your verified role is shown, never your employer unless you choose.
Sign in with LinkedIn to answer
Read next
Chaos Predictions — every Monday
Where enterprise AI is heading next, read from the week’s deployments across banking, pharma, manufacturing and retail.
Free. One email every Monday. Unsubscribe in one click.












