Developing mental health AI tools that improve care across different groups and contexts
In order to realize the potential of mental health AI applications to deliver improved care, a multipronged approach is needed, including representative AI datasets, research practices that reflect and anticipate potential sources of bias, stakeholder engagement, and equitable design practices.
Related Publications
In January 2026, Utah announced a first-of-its kind pilot program allowing an autonomous artificial intelligence (AI) agent to renew prescriptions for consumers who request it. The state agreed not to enforce its unprofessional conduct laws against the developer, Doctronic, if the company adheres to a contract that includes safety and privacy protections. The pilot program includes 192 drugs for chronic conditions. Although physicians will initially validate the AI’s actions, the pilot program will swiftly become one of the first deployments at scale of an autonomous, agentic system in medicine. The announcement prompted concern from associations of physicians and pharmacists who opined that AI “should NOT be making care decisions.”
In January 2026, Utah announced a first-of-its kind pilot program allowing an autonomous artificial intelligence (AI) agent to renew prescriptions for consumers who request it. The state agreed not to enforce its unprofessional conduct laws against the developer, Doctronic, if the company adheres to a contract that includes safety and privacy protections. The pilot program includes 192 drugs for chronic conditions. Although physicians will initially validate the AI’s actions, the pilot program will swiftly become one of the first deployments at scale of an autonomous, agentic system in medicine. The announcement prompted concern from associations of physicians and pharmacists who opined that AI “should NOT be making care decisions.”
The AI Arms Race In Health Insurance Utilization Review: Promises Of Efficiency And Risks Of Supercharged Flaws
Health insurers and health care provider organizations are increasingly using artificial intelligence (AI) tools in prior authorization and claims processes. AI offers many potential benefits, but its adoption has raised concerns about the role of the “humans in the loop,” users’ understanding of AI, opacity of algorithmic determinations, underperformance in certain tasks, automation bias, and unintended social consequences. To date, institutional governance by insurers and providers has not fully met the challenge of ensuring responsible use. However, several steps could be taken to help realize the benefits of AI use while minimizing risks. Drawing on empirical work on AI use and our own ethical assessments of provider-facing tools as part of the AI governance process at Stanford Health Care, we examine why utilization review has attracted so much AI innovation and why it is challenging to ensure responsible use of AI. We conclude with several steps that could be taken to help realize the benefits of AI use while minimizing risks.
Health insurers and health care provider organizations are increasingly using artificial intelligence (AI) tools in prior authorization and claims processes. AI offers many potential benefits, but its adoption has raised concerns about the role of the “humans in the loop,” users’ understanding of AI, opacity of algorithmic determinations, underperformance in certain tasks, automation bias, and unintended social consequences. To date, institutional governance by insurers and providers has not fully met the challenge of ensuring responsible use. However, several steps could be taken to help realize the benefits of AI use while minimizing risks. Drawing on empirical work on AI use and our own ethical assessments of provider-facing tools as part of the AI governance process at Stanford Health Care, we examine why utilization review has attracted so much AI innovation and why it is challenging to ensure responsible use of AI. We conclude with several steps that could be taken to help realize the benefits of AI use while minimizing risks.
AI, Health, and Health Care Today and Tomorrow: The JAMA Summit Report on Artificial Intelligence
Automated real-time assessment of intracranial hemorrhage detection AI using an ensembled monitoring model (EMM)
Artificial intelligence (AI) tools for radiology are commonly unmonitored once deployed. The lack of real-time case-by-case assessments of AI prediction confidence requires users to independently distinguish between trustworthy and unreliable AI predictions, which increases cognitive burden, reduces productivity, and potentially leads to misdiagnoses. To address these challenges, we introduce Ensembled Monitoring Model (EMM), a framework inspired by clinical consensus practices using multiple expert reviews. Designed specifically for black-box commercial AI products, EMM operates independently without requiring access to internal AI components or intermediate outputs, while still providing robust confidence measurements. Using intracranial hemorrhage detection as our test case on a large, diverse dataset of 2919 studies, we demonstrate that EMM can successfully categorize confidence in the AI-generated prediction, suggest appropriate actions, and help physicians recognize low confidence scenarios, ultimately reducing cognitive burden. Importantly, we provide key technical considerations and best practices for successfully translating EMM into clinical settings.
Artificial intelligence (AI) tools for radiology are commonly unmonitored once deployed. The lack of real-time case-by-case assessments of AI prediction confidence requires users to independently distinguish between trustworthy and unreliable AI predictions, which increases cognitive burden, reduces productivity, and potentially leads to misdiagnoses. To address these challenges, we introduce Ensembled Monitoring Model (EMM), a framework inspired by clinical consensus practices using multiple expert reviews. Designed specifically for black-box commercial AI products, EMM operates independently without requiring access to internal AI components or intermediate outputs, while still providing robust confidence measurements. Using intracranial hemorrhage detection as our test case on a large, diverse dataset of 2919 studies, we demonstrate that EMM can successfully categorize confidence in the AI-generated prediction, suggest appropriate actions, and help physicians recognize low confidence scenarios, ultimately reducing cognitive burden. Importantly, we provide key technical considerations and best practices for successfully translating EMM into clinical settings.