Oversight asymmetry in AI-assisted clinical documentation: a comparative-abductive study using neonatal discharge as a test context

Background Human oversight is widely treated as a safeguard for clinical artificial intelligence (AI), yet a reviewer may remain formally responsible without having the knowledge, time, or authority required to verify generated documentation. We develop “oversight asymmetry” as a prospective governance diagnosis and use neonatal discharge as a theoretical test context for responsible digital paediatrics, not as an observed setting of AI use. Methods We conducted a comparative-abductive, netnographically informed qualitative study of 64 public Reddit threads posted between 2021 and April 2026, with comment trees retrieved under pre-specified extraction parameters. The threads contained 2,848 comments and replies from 1,723 pseudonymised accounts in nursing, neonatal, paediatric, medicine, and health informatics communities, and 49 threads contributed to the final themes. A discharge and transition-home stream, drawn mainly from general nursing communities, and a smaller AI documentation and communication stream were analysed separately using reflexive thematic analysis before abductive comparison. Professional role was attributed only where contributors explicitly self-positioned. Results Discharge readiness was described as depending on knowledge distributed across records, professionals, bedside observation, and families, while responsibility for release could become concentrated. Frontline responsibility was separable from institutional authority, although several accounts described distributed oversight that worked because escalation was available. In four anchor threads from two communities, AI documentation discussions described faster drafting alongside residual source checking and correction, and a narrower set of accounts described reconstruction of missing clinical rationale. Workload accounts, together with AI-linked predictions in two threads, suggest that anticipated efficiency may be converted into higher workload before residual verification work is removed. We propose that oversight becomes vulnerable when verification responsibility and demand exceed epistemic access, protected time or consequential authority. Conclusion Oversight asymmetry identifies a prospective mechanism through which nominal human review may fail to constitute meaningful oversight. Responsible paediatric artificial intelligence governance should measure residual verification work, preserve provenance and access to family and bedside context, protect review time, and provide effective output-level and system-level escalation authority before automation is treated as additional capacity. The study generates propositions for prospective evaluation and does not claim that artificial intelligence-assisted neonatal discharge was directly observed.