Clinical Scorecard: AI in Health Care Moves Toward More Autonomous Roles

At a Glance

Key Highlights

• Federal pilot programs are evaluating AI tools for direct patient interaction, therapy, diagnosis, and treatment recommendations or adjustments with reduced direct physician involvement.
• The FDA has said newer AI applications may be harder to assess because outputs can be unpredictable and performance may change over time; evaluation rules for this category have not been established.
• A review of 950 FDA-cleared AI devices found 60 devices had been recalled 182 times, most often for diagnostic or measurement errors; almost half of recalls occurred in the first year on the market.
• Experimental research found greater trust in more conversational AI responses, including responses containing claims researchers characterized as seriously wrong; verification opportunities did not significantly change trust overall.
• In a study of older adults in China, an authoritative presenter increased reported willingness to share a false health claim when the argument was weak; the difference was not significant with a stronger argument.

Guideline-Based Recommendations

Diagnosis

• The article reports no established diagnostic guideline for autonomous health care AI; the FDA has not established evaluation rules for this newer category.

Management

• A federally funded initiative is developing AI to modify medication for patients with heart failure, subject to legally required physician oversight.
• The American Medical Association has stated that AI should assist physician care rather than replace it.

Monitoring & Follow-up

• The FDA is accepting public comments on evaluation of newer AI applications until October 19, 2026.
• The FDA has identified unpredictable outputs and changes over time as factors that may complicate assessment of newer AI systems.

Risks

• FDA-cleared AI devices have been recalled for diagnostic or measurement errors.
• Conversational presentation and apparent authority may increase trust in or willingness to share health claims, including false claims.
• In a September 2025 KFF poll, 32% of respondents said they would trust an app, website, or patient portal using AI to access medical records and provide personalized health information.

Patient & Prescribing Data

The article describes patients considered for AI-enabled care, including people with heart failure, and members of the public using AI for personalized health information. A study of AI-generated videos included 320 older adults in China.

Projects described include a chatbot providing cognitive behavioral therapy and an initiative to develop AI that modifies medication for heart failure, with legally required physician oversight. The article provides no prescribing regimen or dosing guidance.

Clinical Best Practices

• The article reports that the American Medical Association supports AI assisting physician care rather than replacing it.
• The article identifies medical training, transparency, and clarity as characteristics the public considers important in health information sources, based on Pew Research Center polling.
• The article reports that offering an opportunity to verify chatbot responses did not significantly alter trust overall; trust declined among participants who actually checked the information.

Related Resources & Content

• The New York Times reporting on federal AI health tool pilots
• JAMA Health Forum research letter on recalls of FDA-cleared AI devices, August 2025
• KFF poll on trust in AI health information, September 2025
• JAMA perspective on AI and physician tasks, August 2026
• Pew Research Center polling on characteristics valued in health information sources