Divergent responses to AI-era submission pressure: a comparative analysis of medicine and computer science, 2015–2026

ChatGPT's release in late 2022 and the arrival of agentic AI systems from 2025 coincided with reports of rising manuscript volumes and editorial strain. Whether this pressure is uniform across research fields, or absorbed differently by each, has not been tested comparatively. We contrast two ecosystems with sharply different barriers to entry: computer science (CS), where submission carries almost no procedural cost, and medicine, where ethics review and data governance constrain what can be produced. Using open data for 2015–2026 — monthly submission and output volumes from arXiv, medRxiv, bioRxiv and PubMed; submission and acceptance figures for thirteen CS conferences; and review times for 38 medical journals — we measured AI-associated writing with a 95-word excess-vocabulary detector applied identically to 30,000 CS abstracts and 13,000–24,000 PubMed abstracts per year. CS preprint submissions rose 99.3% between 2022 and 2025, against 38–41% for the life-science preprint servers and 5.5% for PubMed-indexed output, and several medical specialties contracted; the divergence holds within arXiv itself. CS conferences absorbed the surge by expanding capacity, with NeurIPS submissions up 195% at an unchanged acceptance rate. AI-associated writing rose sharply from 2023 in both fields, the proportional increase nearly identical from a higher CS baseline, with the break appearing in all eight medical specialties tested. Review time for accepted papers showed no aggregate change. Medicine and CS therefore experienced comparable growth in AI-associated writing but opposite responses: CS expanded output and review capacity, while biomedicine held output flat and routed the pressure into editorial triage.