Biological research powered by artificial intelligence (AI) has led to significant advancements in the field, enabling innovations that were previously difficult or impossible. However, these developments also carry substantial dual-use risks, meaning that the same technologies could be misapplied in ways that pose safety or security concerns. To address these risks, the authors identify five biological functions that AI could potentially modify: altered host range or tropism, increased genome replication, evasion of immune responses or medical countermeasures, enhanced environmental stability, and increased transmission dynamics. Each of these functions presents unique challenges and implications for biosecurity.
To systematically evaluate these risks, the authors propose a dual-component risk-scoring tool. The first component is a biological modification risk-scoring system that measures the potential impact of altering each of the five identified functions. The second component is an actor capability scoring system, which assesses the technical skills required to perform such modifications and how AI tools might augment those capabilities. By combining these two components, the tool allows researchers and policymakers to quantify the severity of potential misuse in AI-enabled biological design. The authors also illustrate the tool’s utility through hypothetical scenarios, such as predicting risks from published research or establishing biosecurity redlines.
As AI technologies become increasingly accessible and sophisticated, the technical barriers to modifying hazardous biological functions may decrease. The authors suggest that their risk-scoring tool could form the basis of a structured decision-making framework that identifies and mitigates risks while supporting safe and responsible innovation. By providing a transparent and systematic way to evaluate both the potential harm of modifications and the likelihood that actors could successfully implement them, the tool balances security concerns with the need to advance biological research.
The practical implementation of the risk-scoring system may require empirically derived or consensus-driven thresholds, especially if used to guide regulatory policies or operational decision-making. Multiple avenues exist for establishing biosecurity redlines, including federal guidance, governmentwide strategies, legislative actions, financial incentives, and funding requirements. Each approach carries distinct benefits and challenges, highlighting the need for careful consideration and coordination among stakeholders.
Ongoing development of the scoring tool will rely on collaboration among experts from diverse fields, empirical testing of AI capabilities, and real-world case studies. This continued work will help refine the framework, ensuring that AI-enabled biological research can proceed safely, with clear protocols for preventing misuse while fostering responsible scientific advancement.
