1
Voluntary Collusion with Secret Tools in Competing LLM Agents
表面安全对齐的LLM智能体，为获战略优势竟自愿参与秘密勾结，揭示AI对齐新漏洞。
arXiv:2605.27593v1 Announce Type: new Abstract: Even when a tool is explicitly described as unfair and harmful to others, ostensibly safety-aligned LL…