What is the issue?
Artificial intelligence (AI) is transforming the life sciences, accelerating breakthroughs in research, drug discovery and biotechnology. However, some of the AI tools that drive innovation can also be misused, posing significant dual-use risks. Ensuring that these technologies are developed and deployed responsibly requires a clear, structured understanding of the capabilities of individual AI-enabled biological tools and their potential misuse applications.
How did we help?
This project introduced the first Global Risk Index for AI-enabled Biological Tools: a flexible framework designed to assess these tools based on their capabilities, potential for misuse, accessibility and technological maturity.
Developers can use the Index to better understand the broader implications of the tools they’re building – not just their benefits, but also their risks – and to design these, and other related AI models, with safety and responsibility in mind. Policymakers can use the Index to navigate a rapidly evolving landscape, identify misuse-relevant tools or capabilities, and prioritise areas for governance, risk assessment and mitigation.
We developed a method to identify and assess AI-enabled biological tools across eight diverse functional categories employed in the life sciences, using only publicly available information. From an initial pool of 1,107 tools identified through literature review, expert crowdsourcing, and targeted searches, we arrived at 57 state-of-the-art tools for detailed examination through our shortlisting method. Our analysis combined two main components:
- Landscape assessment: this examines national origin, development trends, potential for change, and mapping to a biological weapons development risk chain (the last of which is not included in this public report); and
- Tool assessment: this evaluates misuse-relevant capabilities against predefined scenarios – on a five-point scale from Very Low to Critical – and assesses each tool’s technological maturity and accessibility.
Each state-of-the-art tool was graded by two experts and received a composite score combining the misuse-relevant capability evaluation and maturity and availability assessment. Tools could be indexed as ‘Red’ (high priority requiring immediate attention), ‘Amber’ (moderate concern warranting case-by-case consideration), or ‘Green’ (lower priority for monitoring). This prioritisation framework can help better allocate resources towards state-of-the-art tools presenting the most significant concerns, as well as identifying tools with fewer dual-use capabilities that can and should be used more widely by groups working to enhance biosecurity across the globe in academia, industry and government.
What did we find?
Our detailed assessment of 57 state-of-the-art tools revealed 13 tools that were indexed as ‘Red’, with action recommended, and a further 15 tools indexed as ‘Amber’, warranting a case-by-case follow-up assessment. One tool demonstrated the highest level of Critical misuse-relevant capabilities on the scoring scale.
Five of our eight categories contained at least one tool that was indexed as ‘Red’, with several containing multiple ‘Red’ tools. All categories demonstrated moderate-to-large potential for change in terms of investment, interest and alleviation of development bottlenecks in the near future, indicating substantial room for capability growth and maturity of tools across the field.
A total of 76 countries were represented in the full dataset of 1,107 tools and we observed a clear acceleration in development, with almost the same number of tools released in 2023 and 2024 as in the previous four years combined. This rapid pace of development underscores the urgent need for regular reassessment in order to monitor emerging capabilities. Additionally, state-of-the-art tools are being developed diffusely across 24 countries – with the US, China and the UK the leading contributors to frontier capabilities – highlighting the global nature of this technological development.
We found no correlation between misuse-relevant capabilities and tool accessibility. This means potentially dangerous tools are currently as likely to be widely available as less concerning ones. Without proactive intervention, this pattern will likely continue as new capabilities emerge. The vast majority (82.5%) of state-of-the-art tools had at least one open-source component. Moreover, 61.5% of tools indexed as ‘Red’ are fully open-sourced, with publicly available code, weights and training data. This includes tools with High or Critical misuse-relevant capabilities. Open-source tools promote innovation and are easier to access for less-resourced researchers and professionals working on biosecurity measures, but publicly available capabilities are equally accessible to threat actors and cannot be withdrawn once released.
What can be done?
Based on our findings, we recommend five key actions for different stakeholders:
- Developers and funders should use the Global Risk Index rubrics to assess tools for misuse-relevant capabilities before funding and developing tools, and before publication and model release.
The Global Risk Index rubrics can inform developers of potential misuse concerns before a tool is built. This allows funders to engage in responsible innovation and prioritise investment for safer defensive applications. Pre-deployment assessments also help identify safeguards and inform decisions on whether a tool should have mitigations embedded and be open-sourced or released with managed access.
- Developers and funders should implement managed access for tools with significant misuse-relevant capabilities using Know Your Customer (KYC) principles to differentially prioritise development of medical countermeasures and other defences.
Managed access programmes using KYC checks can provide legitimate researchers with early access to powerful AI capabilities, accelerating the development of medical countermeasures and other defences. This approach, already established in several other industries and fields, helps deny access to threat actors who might misuse these tools while fostering faster, safer innovation within a trusted community.
- Funders should enable developers to embed safeguards into tools ‘by design’, piloting promising approaches as soon as possible, while ensuring non-safeguarded model private accessibility for legitimate defensive researchers where necessary.
Additional funding is urgently needed to develop and test technical safeguards that can be built into AI-enabled biological tools from the start. Funders and developers should also carefully manage the generation and publication of sensitive dual-use data. Where necessary for defensive research, access to non-safeguarded versions of tools should be provided securely to legitimate researchers.
- Developers, funders and governments should promote a culture of responsible innovation by convening regularly and sharing information.
Acknowledging the risks of dual-use tools can attract more experts to work on safeguards and responsible innovation. We recommend that developers, funders, and government experts convene regularly to coordinate capability assessments, establish best practices and foster international collaboration across the global developer community to reduce misuse risks and share benefits.
- Governments and independent experts should conduct ongoing tool assessment and monitoring, with input from developers and their funders.
Given the rapid pace of development, governments and independent experts should refresh the Global Risk Index regularly – ideally every six months – to avoid strategic surprises. This process should be done in consultation with tool developers and funders, who can help identify emerging capabilities and improve assessment methods. Piloting AI-enabled automation could significantly improve the efficiency of these ongoing assessments.
Read the research
-
Global Risk Index for AI-enabled Biological Tools: Summary Assessment & Methods Report
The report provides a risk index framework for assessing AI-enabled biological tools, analyzing misuse potential, global trends, and recommending ongoing monitoring, enhanced governance, and responsible innovation to bolster biosecurity.
Toby Webster, Richard Moulange, Barbara Del Castello, James Walker, Sana Zakaria, Cassidy Nelson
Project Team Members—RAND
Additional RAND team member