Skin-on-a-Chip for Cosmetic Safety Assessment: From Reconstructed Skin Models to Regulatory Translation

Cosmetic safety assessment increasingly relies on human-relevant nonanimal methods that support mechanistic interpretation and align with a defined decision context. Reconstructed human epidermis (RhE) supports accepted irritation and corrosion procedures. Reconstructed human skin (RhS) and full-thickness equivalents provide tissue-level comparators for dermal architecture, penetration, and recovery. These models are distinct comparator classes. Rather than cataloging skin-on-a-chip (SoC) architectures, this review assigns an appropriate comparator to each cosmetic-safety question. It then defines the residual uncertainty that may justify SoC testing. It also specifies required readouts and evidence thresholds for a prespecified decision. Vascular and immune modules, controlled topical dosing, and multiorgan coupling are considered only when they address unresolved time course, exposure route, tissue context, or mechanism. Decision-ready readouts require traceable calibration, explicit units, prespecified normalization, drift control, variability estimates, comparator agreement, and quantitative acceptance criteria. No reviewed skin SoC has broad regulatory acceptance for routine cosmetic-safety decisions. Translation therefore requires four distinct stages: analytical validation, biological validation, context-of-use qualification, and regulatory acceptance. Current evidence supports SoC mainly as a higher-tier complement when established methods leave a decision-relevant uncertainty.