能力、阈值和干预措施

AI and the Evolution of BiologicalNational Security Risks Capabilities, Thresholds, and Interventions Bill Drexel and Caleb Withers About the Artificial IntelligenceSafety & Stability Project About the Authors Bill Drexelis a fellow for the Technologyand National Security Program at the Centerfor a New American Security (CNAS). Hiswork focuses on Sino-American competition,artificial intelligence, and technology asan element of American grand strategy. The CNAS AI Safety & Stability Project is a multiyear,multiprogram effort that addresses the established andemerging risks associated with artificial intelligence. Thework is focused on anticipating and mitigating catastrophicAI failures, improving the U.S. Department of Defense’sprocesses for AI testing and evaluation, understandingand shaping opportunities for compute governance,understanding Chinese decision-making on AI and stability,and understanding Russian decision-making on AI andstability. Previously, Drexel worked on humanitarian innovation at theUN (International Organization for Migration) and on Indo-Pacific affairs at the American Enterprise Institute. Alwaysseeking on-the-ground exposure, Drexel has served as arescue boat driver during Libya’s migration crisis; conductedinvestigative research in the surveillance state of Xinjiang,China; and supported humanitarian data efforts acrosswartime Ukraine. He holds a BA from Yale University andmaster’s degrees from Cambridge and Tsinghua universities. Acknowledgments The authors are grateful to Dr. Sonia Ben Ouagrham-Gormley and Dr. Nathan Paxton for their valuable feedbackand suggestions on earlier drafts of this report, and toMichael Aird, Samuel Curtis, and Christian Ruhl, who allprovided further useful inputs. This report would nothave been possible without contributions from our CNAScolleagues, including Paul Scharre, Vivek Chilukuri, MelodyCook, Rin Rothback, Allison Francis, Jake Penders, Tim Fist,Josh Wallin, Michael Depp, and Noah Greene. The reportwas made possible with the generous support of OpenPhilanthropy. Caleb Withersis a research assistant for theTechnology and National Security Programat CNAS. Before CNAS, he worked as apolicy analyst for a variety of New Zealandgovernment departments. He holds anMA in security studies from GeorgetownUniversity with a concentration in technology and security, and a bachelor’s of commerce from Victoria Universityof Wellington with majors in economics and informationsystems. As a research and policy institution committed to thehighest standards of organizational, intellectual, andpersonal integrity, CNAS maintains strict intellectualindependence and sole editorial direction and control overits ideas, projects, publications, events, and other researchactivities. CNAS does not take institutional positions onpolicy issues, and the content of CNAS publications reflectsthe views of their authors alone. In keeping with its missionand values, CNAS does not engage in lobbying activity andcomplies fully with all applicable federal, state, and locallaws. CNAS will not engage in any representational activitiesor advocacy on behalf of any entities or interests and, tothe extent that the Center accepts funding from non-U.S.sources, its activities will be limited to bona fide scholastic,academic, and research-related activities, consistent withapplicable federal law. The Center publicly acknowledges onits website annually all donors who contribute. About the Technology & NationalSecurity Program The CNAS Technology and National Security Programexplores the policy challenges associated with emergingtechnologies. A key focus of the program is bringingtogether the technology and policy communities tobetter understand these challenges and together developsolutions. 01Executive Summary03Introduction04The Current State ofCatastrophic Biological Risks13AI Safety and Biosecurity25Capabilities to Monitor26Recommendations30Conclusion AI and the Evolution of Biological National Security Risks: Capabilities, Thresholds, and Interventions Executive Summary Today, fast-moving advancements in biotechnology—independent of AI developments—are changing many ofthese risks. A combination of new gene editing tech-niques, gene sequencing methods, and DNA synthesistools is opening a new world of possibilities in syntheticbiology for greater precision in genetic manipulationand, with it, a new world of risks from the developmentof powerful bioweapons and biological accidents alike.Cloud labs, which conduct experiments on others’behalf, could enable nonstate actors by allowing them tooutsource some of the experimental expertise that hashistorically acted as a barrier to dangerous uses. Thoughmost cloud labs screen orders for malicious activity,not all do, and the constellation of existing bioweaponsnorms, conventions, and safeguards leaves open a rangeof pathways for bad actors to make significant progress inacquiring viable bioweapons.But experts’ opinions on the overall s