Table 6. Barriers to safe AI integration in neonatology and proposed mitigations.

AUROC: Area under the receiver operating characteristic curve; CE: Conformité Européenne; EHR: Electronic health record; FDA: US Food and Drug Administration; FHIR: Fast Healthcare Interoperability Resources; GIRISH: Goal, Input, Role, Iterative refinement, Safety verification, and Human accountability; HL7: Health Level Seven; IMDRF: International Medical Device Regulators Forum; LLM: Large language model; LMIC: Low- and middle-income country.

Domain	Specific barrier	Proposed mitigation
Data infrastructure	EHR non-interoperability; data silos; LMIC underrepresentation in training data	HL7 FHIR standards; Vermont Oxford Network data sharing; federated learning; globally representative collection mandates
Algorithmic quality	Class imbalance; unreported calibration; poor generalizability to the local population	Multicenter consortia; external-validation mandates; standardized reporting (Brier score with AUROC); post-deployment monitoring
Regulatory	Undefined pathways for adaptive AI; approval slow relative to model improvement	FDA/CE adaptive frameworks; pre- and post-market reporting; IMDRF harmonization; explicit mapping of clinical AI use to the FDA AI/ML SaMD Action Plan [28] and EU AI Act high-risk requirements [29]
Clinician adoption	Limited AI literacy; alert fatigue; no AI training in fellowship curricula; liability concerns	NeonatAIlogy curriculum; GIRISH as an accessible entry point; co-design with bedside clinicians; transparent performance dashboards
LLM-specific risks	Hallucination; no real-time integration; confidentiality risk with unstructured prompts	Mandatory GIRISH deployment; red-flag training (Table 2, S step); enterprise-only platforms; MV-GIRISH for resource-limited settings; accountability- and verification-based mitigation of automation bias [19]
Equity and access	Performance disparity across populations; enterprise-LLM cost in LMICs	LMIC-representative training; open-source development; low-bandwidth deployment; MV-GIRISH