安全研究人员展示了如何透过在共享联络人或定位中隐藏指令来颠覆广泛使用的AI代理OpenClaw，这表明企业不能仅靠强化现有系统来确保安全，而必须重新设计整个安全架构。

AI代理需要存取企业档案和资料库的权限才能发挥潜力，这带来了开放与安全的两难。这正如中世纪的拉古萨共和国，为了在鼠疫期间维持贸易而发明了检疫隔离制度，而非仅仅加强原有的港口防备。

AI带来的真正挑战在于组织变革而非技术本身。预计未来将有大量企业应用程式内建AI代理，但只有那些愿意进行繁琐的组织与流程转型、重建安全防护机制的企业，才能真正获得AI带来的生产力提升。

Security researchers demonstrated how to subvert OpenClaw, a widely used AI agent, by hiding instructions in shared contacts or pinned locations, illustrating that businesses cannot secure these systems by merely hardening existing processes but must redesign their security architecture.

AI agents require access to a firm's files and databases to fulfill their promise, creating a dilemma where openness is both an enabler and a threat. This mirrors the medieval city-state of Ragusa, which invented the quarantine system to keep its ports open for trade during the plague rather than just increasing guards.

The true barrier to AI adoption is organizational rather than technological. As task-specific agents are rapidly integrated into enterprise applications, only companies that undergo the complex transformation of their processes and security practices will successfully harvest the productivity gains of AI.