The field of clinical Artificial Intelligence (AI) continues its rapid expansion, with new models demonstrating significant potential across various medical disciplines. Researchers at the University of Michigan Medicine have developed an AI algorithm to predict in-hospital mortality and complications following percutaneous coronary intervention, aiming to enhance decision-making in cardiology. Similarly, scientists at the University of California San Diego have created an AI model to predict chemotherapy resistance, enabling more individualized cancer treatment, while Paige has introduced an AI application capable of detecting cancer across more than 17 tissue types. In cardiovascular care, an AI-powered tool from Children’s National Hospital promises earlier diagnosis of rheumatic heart disease, potentially averting surgery for a condition affecting hundreds of thousands globally. These innovations underscore AI's growing role in improving diagnostics and treatment planning. Alongside these advancements, the ethical and safety implications of AI in healthcare are drawing increased attention. The World Health Organization (WHO) has released comprehensive guidance on large multimodal models (LMMs), providing 40 recommendations to ensure responsible deployment, emphasizing safety monitoring, equity, and human oversight. An expert consensus published in JAMIA proposes a framework for safe AI-enabled clinical decision support, advocating for systematic validation, a national safety reporting clearinghouse, and standardized documentation to mitigate risks like automation bias and opaque algorithms. These initiatives reflect a growing consensus on the need for proactive measures to ensure AI tools are trustworthy and reduce patient harm. Bias and transparency remain critical concerns. A review in the Journal of General Internal Medicine highlights how unrepresentative training data and a lack of equity auditing in AI-driven clinical decision support tools perpetuate healthcare disparities, calling for standardized subgroup performance reporting and mandatory fairness assessments. This sentiment is echoed by a systematic review in the Journal of Medical Internet Research, which identifies transparency, validation, and ethical concerns as key factors shaping clinicians’ trust in AI systems, noting that opaque algorithms and perceived bias can hinder adoption. A study in Kenyan primary care further illustrated these challenges, where an LLM-based CDS, while often providing appropriate guidance, occasionally recommended unavailable tests and medications, underscoring the need for local validation and bias monitoring. Regulatory frameworks are actively evolving to address these complexities. A health policy institute has proposed detailed principles for regulating healthcare AI as medical devices requiring FDA approval, emphasizing clear definitions and empirical criteria for autonomous systems. While the FDA has also issued updated guidance that may loosen oversight for certain AI-enabled clinical decision support tools and wearables, a new report to the FDA, "The Illusion of Safety," strongly advocates for a comprehensive governance framework with extensive post-market monitoring, mandatory updates for model drift, and transparent reporting on training data composition and demographic performance. This suggests a tension between facilitating innovation and ensuring robust safety oversight. State and federal governments are also increasing their focus on health AI policy. ARPA-H has launched the ADVOCATE initiative to develop and deploy the first FDA-authorized agentic AI system for clinical care, including an oversight AI supervisor. Concurrently, over 40 state bills have been introduced across 25 states in 2026 to regulate clinical AI, focusing on clinical oversight, patient notification, and consent. Major radiology societies have issued joint guidance calling for enhanced safety monitoring and closer collaboration among stakeholders for ethical AI integration. These developments signal a multi-layered governance landscape combining experimental FDA-authorized systems with tightening state-level regulatory frameworks. LOG Standards emphasizes that as clinical AI models become more sophisticated and integrated into care pathways, robust governance, transparent development, and continuous post-market surveillance are paramount. The findings regarding bias, trust, and the need for local validation underscore the importance of our accreditation processes, which prioritize equitable performance, clinician-AI collaboration, and patient safety. The increasing use of AI in mental health, with studies showing significant symptom improvements from AI therapy chatbots but also concerns about potential risks and parasocial attachment among young users, further highlights the need for stringent standards, crisis-safety requirements, and careful evaluation before widespread deployment. LOG Standards will continue to monitor these developments to ensure our frameworks support the safe, effective, and ethical adoption of AI in healthcare.
