Global Risk Index for AI-enabled Biological Tools (Public Report)

Abstract

Artificial intelligence (AI) is on track to revolutionise the life sciences, driving rapid advancements in research and innovation.AI-enabled biological 'tools' are delivering capabilities across multiple disciplines, with the potential of these tools put on global display in 2024 with the Nobel Prize in Chemistry awarded for protein structure prediction and computational protein design.While such developments promise to transform biological research and applications, they also present dual-use concerns, given that some capabilities that advance the life sciences could also enable misuse.This risk is amplified as large language and reasoning models lower barriers to accessing sophisticated new tools.Addressing these challenges requires a systematic understanding of tool capabilities and accessibility to prioritise governance efforts. OverviewIn this report, we introduce the Global Risk Index for AI-enabled Biological Tools, which provides a structured and scalable framework for the systematic assessment of publicly documented tools and their misuse potential.Our methodology builds upon previous work by the Centre for Long-Term Resilience and RAND Corporation, with significant updates and expansions that together deliver a more comprehensive evaluation approach. How to navigate this documentThis report was drafted to meet the needs of several stakeholders with varying levels of technical background and time.It can be read in full, but for certain use cases, we recommend the following sections: The Introduction provides a background to AI-enabled biological tools and the risk-assessment approach used in this report. The Methodology provides the eight category definitions-which frame the landscape of tools-and then describe both the landscape assessment and tool assessment that underpin the Global Risk Index. The Results include both a summary of the tool assessment (conducted on 57 state-of-the-art tools, with three examples included in this public report) and the landscape assessment, which analyses the global nature of tool development, rate of model release and potential for change. The recommended Extensions provide an overview of complementary additions for a deeper and longer-term analysis which model developers, funders, governments, academics and thinktanks could consider using to enhance the Global Risk Index. The Recommendations describe five actions relevant to developers, funders and government policymakers that-based on our findings-we believe can help enhance the governance of these tools and strengthen global biosecurity. The three Appendices provide further details on the report:o The Methodology Appendix describes specifically how we conducted our tool and landscape assessments and provides the rubrics we used for tool and landscape assessment.o The Results Appendix includes additional results and the detailed assessments of three example tools.o The Automation Appendix describes our pilot investigation into the use of large language models to automate some aspects of tool assessment. 8Rose, Sophie, et al. 2024."The near-term impact of AI on biological misuse."The Centre for Long-Term Resilience.https://www.longtermresilience.org/reports/the-near-term-impact-

Ähnliche Arbeiten