‘Navigating the AI diagnostic dilemma’ is healthcare’s No. 1 patient safety concern in 2026
“Navigating the AI diagnostic dilemma” is the No. 1 threat to patient safety in 2026, according to a new report published Monday.
Radiologists and other physicians must balance the benefits of using such technology alongside potential risks. As more healthcare organizations rely on AI to interpret imaging and other clinical data, ECRI is urging leaders to take “a balanced approach to adoption.”
The patient safety organization—formerly known as the Emergency Care Research Institute—compiled its latest annual top 10 list based on a wide scope of data. ECRI aimed to pinpoint the “most pressing threats to patient safety” via scientific literature, safety incident reports and other sources.
“Indeed, AI has been successfully adopted in certain diagnostic radiology procedures for years, and studies have shown that AI technology has the potential to improve diagnostic accuracy and timeliness,” ECRI said in its report, published March 9. “In some cases, it has shown improved diagnostic performance compared to doctors. However, AI systems are only as good as the algorithms they use and the data on which they are trained, and the potential for errors remains a significant concern.”
The institute shared a few potential examples, based on previously published literature. For instance, tested machine learning models failed to recognize 66% of critical or deteriorating health conditions and injuries in synthesized cases, according to one analysis. In another, popular generative AI models more accurately diagnosed genetic conditions tied to textbook-like descriptions, but their accuracy dropped precipitously when prompts were based on a conversation with a simulated patient. And due to the lack of robust training data, AI may have a harder time detecting certain cancers or rare diseases in imaging studies, a 2025 MIT Technology Review study found.
Bottom line: AI isn’t foolproof and can contribute to diagnostic mistakes for which radiologists can be held liable. These models also can perpetuate biases, come with a lack of transparency and erode clinicians’ critical thinking skills.
“In order for AI to be used effectively in diagnosis, clinicians must view it as a tool designed to supplement and support clinical expertise—not replace it,” ECRI charged. “This requires a balanced approach to adoption, thoughtfully considering both the benefits and risks of AI to the diagnostic process. Clinicians who want to best utilize an AI system for diagnosis must be trained on the system’s proper use and must understand its capabilities and limitations.”
The institute also offered potential recommendations to help head off these patient safety issues. Actions could include establishing AI deployment policies, ensuring staff are properly trained on its use, carefully evaluating the business case for these diagnostic tools against costs tied to harm, and disclosing their deployment to patients.
Here is ECRI’s full list of top patient safety concerns for 2026, with No. 2 also highlighting imaging:
- Navigating the AI diagnostic dilemma.
- Reduced access to rural healthcare.
- Increasing rates of preventable acute diseases.
- Federal funding cuts hinder healthcare operations and safety.
- Lack of recognition and reporting of harm events.
- Inadequate pain management for women.
- Persistent workforce shortages.
- Culture of blame hinders learning and improvement.
- Emergency department boarding.
- Gaps in manufacturer packaging and labeling undermine medication safety.
You can download the full report for free here.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.