Commercial artificial intelligence (AI) applications in cardiopulmonary imaging have expanded rapidly—723 Food and Drug Administration (FDA)-authorized radiology AI/machine learning (ML) devices had been identified by June 2024, representing 76% of 950 total cleared AI/ML medical devices, yet evidence supporting clinical adoption remains heterogeneous. This state-of-the-art review synthesizes published evidence across four cardiopulmonary subdomains (coronary, cardiac, aortic, and pulmonary artery imaging), applying the Fryback–Thornbury diagnostic efficacy hierarchy as a unified framework for cross-domain comparison, and integrating regulatory, ethical, and equity considerations aligned to the European Union (EU) AI Act (Regulation 2024/1689) and FDA lifecycle management milestones through early 2026. Evidence maturity varies markedly. Computed tomography-derived fractional flow reserve is supported by randomized trials and prospective registry data addressing clinical decision-making (Level 4); preliminary, non-replicated cost data are also available (Level 6), but no trial has demonstrated improvement in major adverse cardiac events (Level 5) as a primary endpoint. Cardiac AI achieves high reproducibility for automated quantification (intraclass correlation coefficient 0.80–0.95) but remains predominantly at Levels 1–2, with a single non-inferiority randomized trial. Aortic AI evidence is almost entirely retrospective (Level 2). Pulmonary embolism triage systems have demonstrated prospective workflow improvements, including reductions in time-to-diagnosis and radiologist miss rates, without demonstrated mortality benefit. Across all domains, only 5% of FDA-cleared devices underwent prospective testing, and only four of 213 (2%) European regulatory-approved radiology AI devices were clearly labeled for pediatric use. Algorithmic bias, explainability limitations, and immature post-market surveillance frameworks represent consistent cross-domain concerns. Realizing patient benefit requires prospective outcome-driven evaluation, representative training datasets, and rigorous post-market monitoring, particularly as EU AI Act high-risk obligations apply from August 2026.