Healthcare does not need AI that sidelines clinicians. It needs AI governance that protects them.

It starts with a simple principle: AI should support clinical judgment, not replace it.

That is the foundation of this AI Decision Support Charter, a practical governance framework designed to protect what I believe is essential in modern medicine:

- Clinical decisions remain the responsibility of licensed professionals - AI recommendations should never automatically trigger care without human evaluation

- Clinicians must retain the right to override, modify, or reject AI outputs without penalty

- Accountability for AI-influenced decisions must be shared across organizations, developers, and vendors, not pushed onto the clinician alone

Patients deserve transparency when AI meaningfully shapes recommendations about their care. In other words, this charter supports the concept of the sovereign clinician.

A sovereign clinician is not anti-AI.A sovereign clinician is a professional whose expertise, judgment, ethics, and relationship with the patient remain primary, even in an AI-enabled environment.

That distinction matters.

Because if governance is weak, clinicians risk becoming downstream signatories for decisions shaped by systems they did not design, cannot fully inspect, and may be pressured to follow. That is not augmentation. That is erosion.

I am sharing this PDF because I believe healthcare needs stronger language, stronger safeguards, and a clearer defense of clinician autonomy before AI becomes further embedded in clinical workflows.This framework is also part of a larger idea explored in my upcoming book, The Sovereign Clinician, which will be going live on Amazon at the end of this month.

If you care about clinical judgment, patient trust, and the future of accountable AI in medicine, I hope you’ll read the attached PDF and tell me what should be added, challenged, or strengthened.

What the AMA Has Already Put in Writing

The charter's second bullet point, that clinicians must retain the right to override, modify, or reject AI outputs without penalty, is not just a good idea I'm proposing. It closely tracks existing American Medical Association policy. AMA Policy H-480.939 states that oversight and regulation of health care AI systems must be based on risk of harm and benefit, and that liability and incentives should be aligned so that whoever is best positioned to know the AI system's risks, and best positioned to avert or mitigate harm, bears that responsibility, not automatically the clinician AMA Policy H-480.939, Augmented Intelligence in Health Care. The AMA goes further in a 2026 resolution: where a mandated use of AI systems prevents mitigation of risk and harm, the individual or entity issuing the mandate must be assigned all applicable liability, and physicians should not be penalized if they choose not to use AI systems while regulatory oversight and clinical validation remain in flux AMA PPPS Resolution 1, AI Scope of Practice.

That last point matters enormously for the sovereign clinician idea. It is not anti-AI to decline a tool that hasn't earned your trust yet. The AMA's own policy protects that choice, at least in principle. Whether your hospital's actual policies and performance incentives protect it in practice is a separate, harder question, and one every governance charter needs to answer explicitly rather than assume.

The Courts Have Not Caught Up, So You Have To

Here is the uncomfortable truth behind the charter's third principle about shared accountability: as of early 2026, there have been no US malpractice jury verdicts in which AI itself was the central basis of a claim. Courts have relied on traditional malpractice principles, and as Boston University law professor Christopher Robertson put it, "the physician is still the focal point of liability. Courts are likely to focus on whether the clinician acted reasonably in relying on the system, questioning it, or overriding it" AI on Trial: Who's Liable When Clinical Algorithms Go Wrong?, Medscape. Johns Hopkins law professor Stacey Lee adds that there is no doctrine assigning shared legal responsibility to the technology itself, even when AI plays a meaningful role in the decision.

That leaves clinicians in a genuinely asymmetric bind, one legal analysis lays out with unusual clarity: physicians can be liable for wrongly following AI, liable for failing to use an available high-accuracy tool, liable when overriding an alert that later proves correct, and liable when following an alert that later proves wrong Clinical Decision Support AI in 2026, LinkedIn. A physician liability insurer put it even more bluntly for anyone still unsure who is on the hook: "The clinician whose name is on the chart is the clinician who bears responsibility for what's in it... There is no federal law that shifts malpractice liability from a clinician to an AI tool or its developer" AI in Clinical Practice: Who Is Liable?, CMF Group.

This is precisely why a governance charter cannot just gesture at "shared accountability" and move on. Until case law or legislation actually redistributes liability, the sovereign clinician carries it, whether the charter says so or not. What a charter can and should do is create the documentation trail that protects a clinician when they exercise independent judgment. One legal framework proposed for evaluating these cases asks courts to weigh three factors: the clinician's understanding of the AI's limitations, whether the AI's recommendation significantly deviated from established guidelines, and whether the physician exercised independent clinical judgment in interpreting it The New Standard of Care? AI and Medical Malpractice Law. Document that reasoning every time, and you are building your own defense in real time.

• Document the clinical reasoning behind every AI override or acceptance, not just the outcome, this is your defensible record if a case is ever reviewed
• Push your hospital's legal and risk management teams to state explicitly, in writing, who assumes liability when AI use is mandated by policy rather than chosen by the clinician
• Track near-miss cases where a clinician's override of AI proved correct; these are your strongest evidence for preserving override rights in contract negotiations with vendors

Automation Bias Is the Real Threat to Clinical Sovereignty

The charter warns that weak governance turns clinicians into "downstream signatories" for decisions they didn't design. There's a name for the psychological mechanism that makes this erosion happen quietly, without anyone deciding it should: automation bias, the tendency to over-trust an automated recommendation. A randomized trial of 44 physicians who had already completed roughly 20 hours of AI-literacy training found that when exposed to deliberately flawed AI advice, their diagnostic-reasoning accuracy fell from 84.9% to 73.3%, even though they had been trained specifically to watch for this What the Evidence Says About Trusting an AI Suggestion Too Much. Training alone does not inoculate against it.

A global Philips survey found the same gap between intention and infrastructure: 90% of healthcare professionals said keeping a human in the loop was essential, and 86% said all AI outputs required human oversight, yet 70% said training for AI-enabled tools at their own organization was unavailable, limited, or inconsistent AI saves clinicians time but most lack training, Reuters. Everyone agrees oversight matters. Almost no one has built the systems that make oversight possible in practice, at the speed a busy shift demands.

This is the real argument for a written charter, not as a PDF to file away, but as an operational habit. A sovereign clinician's independence erodes one small, reasonable-seeming deferral at a time, not in one dramatic override. The charter's job is to build in the friction, the pause, the documented "why," that keeps deference a choice instead of a habit.

Continue the Conversation

If this resonated, here is where to go next: Join the AI-in-Healthcare Workshop · Get the Books · Contact Dan