Health Systems Say They Validate AI but Lack Dedicated Testing Platforms

Overview

• A report from the Center for Connected Medicine at UPMC and KLAS Research released Aug. 6, 2026 found more than 90% of the 27 health systems interviewed have deployed third‑party AI.
• Among the 25 organizations with deployments, 23 reported pre‑launch validation but only 11 said they have a dedicated testing environment or sandbox to test models before rollout.
• Respondents use very different meanings of “validation,” ranging from simple workflow checks to formal accuracy, bias, drift and safety assessments run against test cases or production‑like data.
• Clinical documentation and ambient scribing are the top use cases, followed by revenue cycle/coding, imaging and EHR‑embedded decision support, which concentrates operational and patient safety risk.
• Leaders describe AI strategy as mostly developing, cite limited resources and talent, and warn that vendor‑embedded tools and weak data pipelines make independent testing, monitoring and ROI measurement harder.