ACP issues ethics guidance for AI at point of care

Clinical takeaway: Disclose AI use to patients when it touches their care, and treat AI outputs as just an input to your judgment, never a substitute for it. ACP frames deskilling and algorithmic bias as ethical risks physicians own, not just system problems. 

Physicians did not wait for substantial ethical guidance before adopting AI, and neither did their health systems. Ambient scribes draft notes, chatbots field patient messages, and algorithms read mammograms while the profession's answer to how these tools should be used sits scattered across governance frameworks written for developers, regulators, and institutions. 

The American College of Physicians thinks that gap is a problem worth solving. Existing ethics guidance is largely concentrated on how systems should govern algorithms rather than on what the individual clinician owes the individual patient when AI enters the encounter. Consensus is emerging around broad values like transparency and fairness, but not necessarily around what those values require in an exam room. A new ACP position paper takes that question head on, grounding the ethical use of AI in the patient-physician relationship itself and offering physicians a working framework for the tools already in their hands. 

The paper stakes out four positions. First, AI must be implemented to support the patient-physician relationship, not weaken it. Second, physicians and trainees must recommit to the competencies of caring while building new competencies for AI use. Third, AI used in decision making must respect and promote physician-independent clinical judgment rather than replace it. And fourth, physicians, working with clinics, hospitals, and health systems, must identify and mitigate algorithmic bias, including its social and structural roots. 

On disclosure, the committee lands on the side of transparency. Telling patients does not require documenting every detail of every model, it argues. But the newness of AI and public unease argue for erring toward disclosure, starting with basics like telling a patient an ambient scribe is running and how the data are handled. The paper cites survey data showing nearly 60% of Americans worry that AI use would worsen their relationships with clinicians. Physicians should be ready to explain what an AI system does and whether it applies to the patient in front of them, the same way they already explain whether trial findings apply. 

Deskilling is an ethical issue, not merely an educational one, because patients are harmed when automated systems err and the physician can no longer catch it. Autonomous AI should be rare, confined to low-risk, low-complexity decisions, and always allow a clinician to be brought into the loop. Efficiency gains from AI should buy more quality time with patients, the paper warns, not an expectation that physicians will see more of them. 

On bias, it pushes past simply technical fixes. Real mitigation means moving upstream to health inequity and the social determinants that shaped the data in the first place. Physicians should neither assume all models are biased nor withhold AI from certain groups, which can itself create a feedback loop that degrades performance for those patients. 

The paper was developed by ACP's Ethics, Professionalism and Human Rights Committee using a narrative-style review of peer-reviewed research, commentaries, and organizational guidelines on AI ethics. It was reviewed by the ACP Board of Governors, multiple councils, and outside experts, and approved by the Board of Regents in April 2026. The framework builds on a 2024 ACP policy paper that framed AI as augmented intelligence, a tool to assist clinicians rather than replace them. 

Much of what the paper calls for does not yet exist. Validated AI fairness metrics remain to be developed, and the committee wants them built into clear, standardized labels that summarize a model's basic features for patients and physicians. Evolving federal rules and state laws may hold covered healthcare entities responsible for algorithmic discrimination, sharpening the stakes of the bias positions. With FDA review of AI-enabled devices still quite limited, ACP argues local governance is essential and points to the models many health organizations are already building as the venue where best practices will get worked out. 

"Now is the time for physicians to consider how to harness AI's transformative potential while avoiding patient harm," the authors conclude. "Ethical principles and guideposts of relationality, self-governance, and competence must shape AI's use in medical practice, based in 'a code of ethics and a duty of service that, in medicine, puts patient care above self-interest,' and medicine's societally granted privilege of self-regulation." 

Source: DeCamp M, et al. (2026 Sep 1) Ann Intern Med. Ethics and Professionalism in Artificial Intelligence and Medical Practice: A Position Paper From the American College of Physicians