Answers are grounded in this paper and cite their sources.
Artificial intelligence (AI) is becoming increasingly incorporated in the life sciences. Until recently, the need for data governance was masked by the limits of our human ability to aggregate, synthesise, and draw inferences from data. AI enhances not only speed of research and data processing, but also dual-use risks from training AI models with open access data. Concerns about AI-enabled chemical and biological weapons development become louder while governments, institutions and model developers scramble to mitigate and contain those risks. Data remains a crucial ingredient for AI model advancement, consequentially the lack of data governance of AI in life sciences leaves us collectively and personally vulnerable. Traditional data governance controls storage, transfer and access to datasets, but AI-enabled life science research raises a different issue: which novel capabilities emerge when models aggregate and orchestrate openly available data and tools, arising from individually benign contributions and not covered by current governance. Thus, we highlight questions and ethical considerations on which precautions are required when risks arise through data aggregation and re-use, how responsibility can be distributed, whether governance should focus on data or novel capabilities and if openness should cease to be default. Data governance works upstream, before data is introduced to an AI model. It follows that more life science dual-use data of concern should be held in Trusted Research Environments rather than released openly, and that dual-use data assessments become routine, implemented through research funders as chokepoints, to bridge the legislative gap until international dual-use standards are established.
No takes yet. Share an insight, caveat, or question.
Rothweiler et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: