Dear Editor,
Li’s report presents a range of responses to AI-enabled biological threats, from protein watermarking to wastewater surveillance and vaccine development. Yet these measures address different stages of the threat pathway, and listing them does not establish that they collectively provide effective protection.
The distinction matters. Watermarking may help identify AI-generated proteins, wastewater surveillance may reveal emerging biological signals, and vaccines may mitigate disease after exposure. None, however, necessarily prevents a malicious actor from developing or releasing a biological agent. The central policy gap is therefore not simply a shortage of surveillance or preparedness, but the absence of a clearly articulated pathway linking early warning to timely, enforceable intervention.
This is not an argument for delaying action until risks are fully quantified. The National Academies has recognised both the potential for AI-enabled biological tools to increase biosecurity risks and the need to assess their specific contributions.¹ Meanwhile, the World Health Organization’s framework emphasises shared responsibility across the life sciences,² and recent protein-watermarking research illustrates the promise—and limited scope—of individual technical safeguards.³
Biosecurity strategies should therefore be judged not by how many safeguards they propose, but by whether those safeguards interrupt the progression from harmful capability to real-world harm. Who acts when a credible warning emerges? What intervention follows, and how is its effectiveness measured?
Without answers to these questions, better detection may make threats more visible without making populations safer.
References
1.National Academies of Sciences, Engineering, and Medicine. The Age of AI in the Life Sciences: Benefits and Biosecurity Considerations. National Academies Press; 2025. doi:10.17226/28868.
2.World Health Organization. Global Guidance Framework for the Responsible Use of the Life Sciences: Mitigating Biorisks and Governing Dual-Use Research. WHO; 2022.
3.Stutz D, Cowen-Rivers AI, Ortiz-Jimenez G, et al. Function-preserving watermarking of AI-generated proteins. Nature. Published September 30, 2026. doi:10.1038/s41586-026-10965-y.