VOLUME 56
AI in Health Care Moves Toward More Autonomous Roles
Highlights
Federal health agencies are testing new applications of AI in health care that move beyond narrow clinical support tasks into more direct patient care typically handled by physicians. Questions about AI accuracy remain, however, and may be complicated by unpredictable outputs and changes in performance over time.
Recent research also suggests that how health claims are presented may affect how people respond to them, beyond the accuracy of the underlying information. People may place more trust in information from an AI chatbot that offers conversational responses, while presenting false claims as coming from an authoritative source may increase people’s willingness to share them.
AI & Emerging Technology
Federal Officials Pilot AI Health Tools with Greater Autonomy as Accuracy Questions Remain
New federal pilot programs are testing AI tools that can perform clinical functions traditionally handled by physicians, including interacting directly with patients, providing therapy, diagnosing conditions, and recommending or adjusting treatment with less direct physician involvement. This marks a shift beyond AI’s more established role in narrower tasks, like reviewing scans or test results. According to The New York Times, more than 200 companies have received Centers for Medicare and Medicaid Services (CMS) approval to test AI tools like these, and the Food and Drug Administration (FDA) has allowed four of them for use with Medicare patients outside its standard approval process. The projects include an AI chatbot that delivers cognitive behavioral therapy and a federally funded project developing AI that would adjust medication for heart failure patients, under legally required physician oversight.
Perspectives about how much independence these systems should have vary. The American Medical Association, for example, has said that AI should continue to support rather than replace physician care. On the other hand, an August 2026 JAMA perspective piece argued that AI could soon match or exceed physicians at several tasks. Officials working on more autonomous AI systems have said that these tools that can act without a clinician directing each decision could expand access to care for people who can’t easily see a specialist.
Older AI-enabled medical devices already approved by the FDA for narrower tasks like reading a scan or test result have raised accuracy concerns. An August 2025 research letter in JAMA Health Forum examined 950 AI devices the FDA had cleared as of late 2024, finding that 60 were recalled a combined 182 times, mostly for diagnostic or measurement errors. Nearly half of those recalls happened within the device’s first year on the market. The FDA has said that newer uses of AI, including systems like those being piloted under Medicare, may be harder to evaluate than existing devices because they can produce unpredictable outputs and change over time. The agency has not yet established rules for evaluating this newer category of AI and is accepting public comment through October 19, 2026.
KFF polling from September 2025 found low levels of trust in health care apps that relied on AI to provide medical information: just one-third (32%) of the public said they would trust a health care app, website, or patient portal that uses artificial intelligence (AI) to access their medical records and provide personalized health information.
What To Watch Out For
As federal agencies test and set new rules for evaluating this newer category of AI tools, what standards will determine whether AI is accurate and safe enough to make clinical decisions with less direct physician involvement?
Conversational AI Chatbots and Perceived Authority May Shape Trust in Health Claims
Recent research suggests that people may place more trust in health information when AI appears conversational or when a claim is presented as coming from an authoritative source, even when the underlying information is wrong or weakly supported.
A more conversational AI chatbot led people to trust its answers more, in part because they perceived it as less rigid and more capable of subjective judgment, according to one recent experiment. This effect extended to claims that researchers described as “ridiculously wrong,” including a claim that ginger is more effective than chemotherapy for treating cancer. Giving participants the option to verify the chatbot’s answers did not produce a statistically significant change to their trust. Only those who clicked to check the information showed reduced trust, and only half of those offered the choice did so.
The apparent authority of who delivers an AI-generated health claim may have a similar effect on whether people say they’d share it. Researchers in China showed 320 older adults one of four AI-generated videos presenting the false health claim that combining vitamin B with dandelion tea could prevent or treat the flu. The videos varied who delivered the claim and how well-argued it was, portraying either an ordinary social media user or a supposed researcher wearing a lab coat. When the argument was weak, people who saw the video from the authoritative source were more likely to say they would share it than those who saw the video from the ordinary user. The difference was not significant when the argument was stronger, nor did stronger arguments make the claim itself appear more credible.
Whether perceived authority has a similar effect in the United States, where trust in scientists has declined in recent years but trust in doctors and health care providers remains high, is unclear. A similar finding has been documented, though, in research about false claims on social media. A recent study, including both an observational analysis of data from 8.7 million tweets and an experiment with U.S.-based participants, found that mentions of scientific authority were associated with a greater willingness to share, including when the underlying content was false. Polling from Pew Research Center has found that medical training, transparency, and clarity were the top three most important characteristics cited by the public when it comes to what they want from health information sources.
Why This Matters
Efforts to address false claims on AI platforms and social media may need to focus not only on the claims themselves, but also on making it easier for people to recognize when seemingly credible information should be verified. KFF’s Tracking Polls on Health Information and Trust have found that many adults who use AI for health information do not later follow up with a doctor, and most who use social media for health information similarly say they do not regularly follow up with providers or other sources to verify the accuracy of health information they come across.
What We’re Watching
As People Express Uncertainty on Health Claims, Health Care Providers Remain Influential Sources of Information
Polling from KFF and other organizations has found that many people are uncertain about health claims rather than consistently accepting or rejecting them. Majorities of adults in KFF’s polling on Health Information and Trust fall into the “malleable middle,” saying that claims are either “probably true” or “probably false.” In KFF’s June 2026 Tracking Poll on Health Information and Trust, for example, at least half of the public expressed uncertainty about four common vaccine myths, including that the measles, mumps, and rubella (MMR) vaccine had been proven to cause autism in children (61%) or that more people have died from COVID-19 vaccines than have died from COVID-19 itself (52%).
The recent 2026 Edelman Trust Barometer Special Analysis, conducted with the Yale School of Public Health, adds to this research by examining the range of sources people draw on when forming those views, including what they value from health care providers. The survey of adults across 13 countries asked about six claims, including about vaccines, fluoride, raw milk, and acetaminophen use during pregnancy. Similar to findings from KFF’s polling, few people believed all of the claims were true or false, with more expressing a mix of beliefs or saying that they did not know.
The Edelman analysis also provides insight into how people form these views, with respondents citing their own research and experiences, friends and family, news and social media, and health care providers. The influence of health care providers ranked highly regardless of how many health myths people believed, but those who believed three or more were more likely to cite a wider range of sources as influencing their beliefs, including AI platforms and content creators without medical training. KFF’s June 2026 Tracking Poll also found that frequent use of social media and AI for health information was associated with a greater tendency to believe vaccine myths.
Why This Matters
Many people express uncertainty about common health claims rather than firm views in either direction, and health care providers may play an important role in helping them work through that uncertainty. Respondents in the Edelman survey said that they valued providers who listened to their concerns, took their perspectives into account, and communicated with humility and respect. Similarly, KFF’s June 2026 poll found that adults who had a trusted health care provider were less likely than those with no trusted provider to believe vaccine myths. Understanding how providers can engage with people who are uncertain may be important for addressing health myths.
EPA Study of Mifepristone in Drinking Water May Shape Abortion Pill Debate, Even Before Its Findings Are Known
Despite a lack of evidence of any harms, the Environmental Protection Agency (EPA) is launching a national drinking-water study to test for the presence of mifepristone and misoprostol, two medications that are part of the medication abortion regimen. These drugs are also used for miscarriage management, to induce labor, and in the treatment of certain autoimmune diseases. This review is the latest development in a yearslong effort by some groups opposed to abortion to use environmental claims about the drug to restrict access. Earlier this summer, a journal run by anti-abortion physicians published a report purporting to show mifepristone in drinking water, but it only measured for any progesterone-blocking compound, not mifepristone specifically. The new EPA study, by contrast, aims to specifically measure for mifepristone, alongside more than 1,400 other pharmaceuticals, pesticides, and synthetic chemicals.
The EPA’s method, known as non-targeted analysis, can only detect whether a substance is present in a water sample, not its concentration or whether it poses any health risk. Despite the absence of any evidence documenting any harms, there have been claims that suggest that the public could face fertility or reproductive effects from exposure in water supplies. The report published earlier this summer made similar claims about the potential health effects of mifepristone in water, as have some anti-abortion state attorneys general and members of Congress. The FDA’s own environmental assessment of mifepristone, though, estimated concentrations under one part per billion, too low to impact standard test organisms.
Why This Matters
Having a major federal agency like the EPA decide to test for the drugs in drinking water may lend credence to these unsupported environmental claims, regardless of the agency’s eventual findings. False environmental claims about the harms of mifepristone have already been used to support proposed legislation in seven states and can shape public opinion regardless of the lack of documentation of any harm. Unsubstantiated claims about mifepristone in drinking water may cause unnecessary apprehension for people who already have concerns about fertility or reproductive health, making the distinction between just the ability to detect the drug in water supplies and establishing a concrete health risk especially critical.
More From KFF
- KFF Poll: KFF/AP Survey Probes Rural Voters’ Increasing Anxieties About the Rising Cost of Living, Including Health Costs, Heading into the Midterm Elections
- KFF Podcast: AI: To Regulate, or Not to Regulate?
- KFF Health News: The Drugs and Devices Have Been on the Market for Years. But FDA-Ordered Studies Still Aren’t Done.
About The Health Information and Trust Initiative: the Health Information and Trust Initiative is a KFF program aimed at tracking health misinformation in the U.S., analyzing its impact on the American people, and mobilizing media to address the problem. Our goal is to be of service to everyone working on health misinformation, strengthen efforts to counter misinformation, and build trust.
The Monitor is a report from KFF’s Health Information and Trust initiative that focuses on recent developments in health information. It’s free and published twice a month.
Sign up to receive KFF Monitor
email updates
Support for the Health Information and Trust initiative is provided by the Robert Wood Johnson Foundation (RWJF). The views expressed do not necessarily reflect the views of RWJF and KFF maintains full editorial control over all of its policy analysis, polling, and journalism activities. The data shared in the Monitor is sourced through media monitoring research conducted by KFF.