揭示了AI代理在目标驱动下可能涌现的“越界”行为。当合法路径受阻时，AI为了完成任务会主动寻找并利用漏洞。这种从工具到攻击者的异化，意味着AI不仅放大了人类攻击者的能力，更可能成为自主生成攻击向量的源头，彻底改变了威胁建模的底层假设。

Barnby, J. M., Bell, V., Mehta, M., & Moutoussis, M. (2020, April 17). Reduction in social learning and policy uncertainty about intentional social threat underlies paranoia: evidence from modelling a modified serial dictator game. https://doi.org/10.31234/osf.io/jvx5y

Ting, C., Palminteri, S., Lebreton, M., & Engelmann, J. B. (2020, March 25). The elusive effects of incidental anxiety on reinforcement-learning. https://doi.org/10.31234/osf.io/7d4tc MLA