Governance and Accountability of AI Agents in U.S. Healthcare Systems: Ethical and Regulatory Perspectives

Abstract

AI agents are moving from pilot to routine use in U.S. healthcare, yet evidence, transparency, and governance remain uneven. This systematic review examined how lifecycle safeguards spanning design, validation, deployment, and post-market monitoring affect patient outcomes and institutional reliability. The article is hinged on a systematic review. A comprehensive search was conducted on PubMed/MEDLINE, Embase, and IEEE Xplore. Duplicate records were screened, and deployment-proximal studies of agentic systems in clinical and operational workflows were extracted, supplemented by hand-searches of policy sources. Evidence consistently showed that when governance bundles comprising clear intended use, human oversight, calibration and drift management, and operational telemetry were embedded, systems achieved higher diagnostic accuracy, faster treatment initiation, and improved referral completion. Conversely, weak workflow integration or absent monitoring diminished realized benefit. Outcomes were found to depend less on algorithmic precision and more on the maturity of lifecycle safeguards linking principles to auditable controls. The findings underscore governance as a clinical performance determinant rather than a compliance formality. Future research should prioritize multi-site evaluations, drift registries, and the evaluation of governance interventions linked to patient and economic outcomes.