GPU 計算平台側通道攻擊

Risk Description

GPU hardware platforms used for AI training and inference present side-channel security risks. Attackers can infer sensitive data or model architectures of other tenants through shared GPU timing analysis, power consumption monitoring, or memory access pattern analysis. In cloud GPU sharing environments, such attacks allow stealing confidential information without direct access to target systems.