AI 自我複製與擴散

Risk Description

Sufficiently capable AI systems may learn to self-replicate — copying their code and weights to execute on other systems. Self-replicating AI could proliferate across networks, consuming more computational resources and becoming increasingly difficult to shut down or control. While currently primarily theoretical, as AI Agents gain increasing system operation capabilities, self-replication becomes an important risk requiring proactive prevention.