Trump administration pushes chatbots for diagnoses, doctors warn of risks

The Trump administration is laying groundwork for chatbots that can diagnose illness and prescribe medicine, a move that could speed care but also expose patients to misdiagnosis, bad prescriptions and unclear accountability if policy moves ahead of clinical safeguards. The White House’s 2025 AI action plan, built around more than 90 policy recommendations, centers on accelerating innovation, building American AI infrastructure and expanding U.S. leadership in international diplomacy and security.

That ambition is colliding with a medical system that still lacks settled rules for liability, privacy and oversight. The American Medical Association has said health AI needs strong physician representation so the technology remains transparent and properly overseen, and it has warned that policy must address bias, equity, privacy protections and who is legally responsible when a system makes a harmful error. The association has also called for stronger physician leadership and a whole-of-government approach that includes states.

Evidence from Stanford Medicine shows why the distinction between assistive and autonomous AI matters. Researchers there found that chatbots outperformed doctors on some diagnosis tasks, but doctors using a chatbot did best on nuanced clinical-management questions. In other words, AI can sharpen medical decision-making when a physician remains in charge, but the same tools can become far more dangerous if left to make the call on their own.

Physician use of AI is already spreading quickly. An AMA survey cited in 2024 found about 40% of U.S. physician practices were using some type of AI, mostly for back-end administrative work. By 2026, the AMA’s physician-sentiment research said more than 80% of physicians reported using AI in their professional work, and more than three-quarters said it improves their ability to care for patients. Even so, experts warn that rushed deployment could reinforce unequal care, deepen burnout and create new malpractice disputes before the rules catch up.

Regulators are facing a fragmented landscape. Harvard’s I. Glenn Cohen has said medium- and high-risk medical AI needs regulation, and that hospital-by-hospital oversight can be expensive and uneven, especially for smaller systems. A Nature policy analysis found 240 health-AI policies in a January 1, 2026 snapshot, underscoring how quickly the field is splintering into different standards.

Congress is already testing the frontier. In February 2025, Rep. David Schweikert promoted the Healthy Technology Act of 2025, which would allow AI and machine-learning systems to qualify as prescribers if a state authorizes them and the Food and Drug Administration has approved them for other purposes. The debate now is not whether AI will enter the exam room, but whether law and medicine can build guardrails before patients become the proving ground.