Safety, Accountability, and Regulation of Artificial Intelligence in Diagnostic Medicine: A Systematic Review Author Artificial intelligence, Diagnostic medicine, Patient safety, Algorithmic bias, Legal accountability, Regulatory oversight. Importance: The Artificial Intelligence (AI) and machine learning (ML) technologies are majorly utilized in radiology and other clinical decision support systems. However, in spite of the regulatory approval of many AI diagnostic systems, their clinical use has progressed before the development of safety frameworks and regulatory standards. Objectives: To review the existing literature on safety, accountability, and the challenges in the regulation of AI diagnostic systems, and the evidence-based measures that could ensure the safe clinical use of the systems. Evidence Review: A literature search was conducted in PubMed/MEDLINE, Embase, Scopus, and Web of Science databases, utilizing artificial Intelligence, machine learning, diagnostic medicine, safety, regulation, accountability, liability, ethics, and implementation as the keywords. Only the studies that covered clinical AI diagnostic systems and analyzed safety, ethics, legal, and regulatory aspects were considered. After a review by two independent reviewers, fifty-three studies were included. The grading of evidence was performed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework. Findings: This systematic review of 53 studies identified empirically documented patient harm arising from algorithmic bias, producing measurable health inequities in deployed commercial systems; explainability deficits that prevented clinicians from exercising independent judgment; legal liability ambiguity that left 89% of jurisdictions without clear accountability frameworks; regulatory oversight failures, as no major jurisdictions had implemented mandatory post-market surveillance for adaptive AI; data privacy violations in cross-border deployments; and implementation barriers that disrupted clinical workflow in 73% of deployment studies. Effective safeguards, including obligatory human-in-the-loop oversight, demographically representative training datasets, specific explainability requirements, explicit institutional liability frameworks, and harmonized post-market surveillance, were identified but implemented inconsistently. Conclusions and Relevance: The AI-assisted diagnosis tools produce documented patient harm through algorithmic bias and unsolved errors. The study found that AI safety is primarily a governance failure and not a technical problem. The compulsory pre-deployment audit of bias, explainability standards modified to clinical stakes, explicit legal accountability frameworks, and harmonized international post-market surveillance are not mere aspirational goals but requirements for preventing foreseeable harm. A clinical deployment without these safeguards constitutes a predictable patient safety risk. UnivColl International Multidisciplinary Research Journal 2026-06-30 Text application/pdf 10.65919/uimrj.2026.v2i6001 en