Dual-Use AI Capabilities and the Risk of Bioterrorism: Converting Capability Evaluations to Risk Assessments

Several frontier AI companies regularly test their models for dual-use biological capabilities that

might be misused by threat actors. But such tests do not tell us the extent to which the overall risk of

bioterrorist attacks would be increased if such tests were to find concerning results – and there is

much expert debate about how seriously to view such threats. This report creates a framework for how

to convert capability evaluations into risk assessments, using a simple model that draws on historical

case studies, expert elicitation, and reference class forecasting. The author concludes that if AI

systems were to increase by 10 percentage points the number of STEM undergraduates able to

synthesize pathogens as complex as influenza and creates significant results in an operational risk

study, then the annual probability of an epidemic caused by a lone-wolf terrorist attack might increase

from 0.1% to 0.6%. This would be equivalent to approximately 10,000 expected deaths per year, or

$120B in damages. Risk scenarios with additional AI virus discovery capabilities reach even higher

damages. A review of this report by subject matter experts and forecasters found similar median

ranges. All forecasts had high uncertainty. This work demonstrates a methodological approach for

converting capability evaluations into risk assessments, whilst also highlighting the continued need

for better underlying evidence and expert discussion to update and narrow our assumptions.

‍