Towards Risk Analysis of the Impact of AI on the Deliberate Biological Threat Landscape

TL;DR

The paper addresses how the convergence of AI and biotechnology could elevate deliberate biosecurity risks and proposes a quantitative framework built on the triplet $S_i$, $P_i$, $C_i$ with AI-augmented analogs $S_{ai}$, $P_{ai}$, $C_{ai}$, coupled to a five-step biorisk chain ($s_i^{idea}$, $s_i^{acq}$, $s_i^{prod}$, $s_i^{weapon}$, $s_i^{deploy}$). It formalizes probability via per-step terms $p_i^{step}$ and $p_{ai}^{step}$ and demonstrates through Monte Carlo analysis—using OpenAI data as proxies for $p^{step}$—that AI can increase overall risk (e.g., from $500$ to $1{,}700$ deaths in a illustrative pair) even when some studies show no statistical difference in $p^{idea}$. The paper also provides a qualitative framework aligned with the NASEM Biodefense in the Age of Synthetic Biology to assess AI impacts across Usability, Weaponization, Actor Requirements, and Mitigation, emphasizing relative risk over absolute quantification. Finally, it advocates keeping evaluation methodologies pace with AI advances, highlights policy relevance to the 2023 Executive Order, and calls for cross-disciplinary collaboration and dedicated tooling to characterize AI-enabled methods in life sciences for risk-informed governance.

Abstract

The perception that the convergence of biological engineering and artificial intelligence (AI) could enable increased biorisk has recently drawn attention to the governance of biotechnology and artificial intelligence. The 2023 Executive Order, Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, requires an assessment of how artificial intelligence can increase biorisk. Within this perspective, quantitative and qualitative frameworks for evaluating biorisk are presented. Both frameworks are exercised using notional scenarios and their benefits and limitations are then discussed. Finally, the perspective concludes by noting that assessment and evaluation methodologies must keep pace with advances of AI in the life sciences.

Figures (1)

• Figure 1: Box-and-whisker plot displaying results of 10,000 Monte Carlo simulations of the probability of successfully developing a plan for the misuse of biology. Data distributions are based on data from OpenAI (tejal_patwardhan_building_2024). Blue results represent participants with access to the internet only. Orange results represent participants with access to the internet and an LLM-based chatbot. Boxes represent the median and inter-quartile range (IQR). Diamonds denote outliers more than 1.5x the IQR from Q1 or Q3. "Overall" results represent the product of the probability of each individual step.