Super Intelligence in Healthcare: A Systematic Review of Systemic Risks, Ethical Dilemmas, and Governance Frameworks

Abstract

Background: Artificial intelligence (AI) is moving from narrow, task-specific tools toward general-purpose systems whose capabilities may eventually exceed expert human performance across most cognitive tasks. Healthcare is among the most consequential domains in which such systems will be deployed, yet governance arrangements remain calibrated to narrow software rather than to open-ended, self-improving agents.

Objective: To synthesise evidence on the systemic risks, ethical dilemmas, and governance frameworks relevant to advanced and prospectively superintelligent AI in healthcare, and to expose the gap between promotional claims and the empirical record.

Methods: A PRISMA 2020-guided structured synthesis was conducted in two streams: (A) a defined corpus of 117 bibliographic records, screened against explicit eligibility criteria, and (B) 51 purposively selected benchmark sources providing comparative empirical, ethical, and regulatory evidence. Findings were organised through thematic synthesis and critical appraisal, and compared across regulatory regimes.

Results: Of the 117 corpus records, 72 were excluded at title screening, 45 were retained, and 33 formed the core synthesis. Four convergent findings emerged. First, the evidence for clinical benefit is dominated by retrospective, single-site, and narrowly benchmarked studies. Second, systemic risks (automation bias, deskilling, correlated failure, security vulnerability, and concentration of power) scale non-linearly with capability. Third, principle-based ethics frameworks are widely endorsed but weakly operational. Fourth, existing regulatory regimes address device performance but not alignment, accountability for autonomous action, or catastrophic tail risk.

Conclusion: Governance of superintelligent healthcare AI cannot be deferred until such systems exist. A staged, evidence-gated framework that couples capability thresholds to oversight obligations, liability allocation, and equity safeguards is proposed.