Parity before scale: governing large language models in global health

Large language models (LLMs) are proposed as scalable solutions for health workforce gaps in low- and middle-income countries, yet deployment risks deepening inequities. LLM evaluation remains disproportionately centered on English-language resources, and current reporting standards do not operationalize linguistic parity, data sovereignty, or accountability for specific language communities, care pathways, and regulatory arrangements. We propose three minimum governance commitments: parity before scale, requiring pre-specified performance thresholds or non-inferiority margins by language and care setting before deployment; local ownership before extraction; and accountability before integration. Systems failing these thresholds should be restricted, redesigned, or subject to additional oversight before scale-up.