Rand兰德：2025年面向前沿大型语言模型生物学知识的综合基准测试研究报告（英文版）

he offense-defense balance for biosecurity is a delicate one that can be disrupted by the introduction of transformative technologies, such as artificial intelligence (AI). Although most public general-purpose AI systems contain safety guardrails that refuse to assist with harmful tasks, these defenses are known to be vulnerable and prone to manipulation. Given the historical interest in biological weapons from malicious actors who lack scientific expertise, we are interested in evaluating wh