A new review found that most AI-enabled medical devices authorized by the FDA had not been tested for patient outcome improvements. Researchers identified 1,357 AI and machine-learning devices cleared for patient care through December 5, 2025 and reported that evidence of clinical benefit via outcomes testing was rare. The findings raise questions about how postmarket performance and real-world effectiveness are assessed for AI systems once they reach clinicians and patients. The analysis also underscores differences between demonstration of technical performance and proof of clinical impact. For biotech and medtech developers, the work points to a potential emphasis shift from model benchmarking toward outcome-driven study designs that can withstand scrutiny in regulatory and payer settings. The review’s conclusions are likely to influence how developers structure submission packages and how stakeholders evaluate whether AI claims translate into measurable patient health improvements.