OWASP publishes top ten security risks for AI data centers; tampered firmware and insufficient high-speed network protection listed as major risks
OWASP公布AI資料中心十大資安風險,韌體遭竄改、高速網路防護不足列重大風險
OWASP released a list of the top ten security risks for new cloud services and AI data centers, five of which are classified as critical risks, focusing on the underlying computing environments that AI model training and inference depend on, covering hardware and firmware, high-speed networks and multi-tenant isolation. The first version of the list was released in July 2026, and OWASP provided further details on November 11. The top five critical risks, in order, are compromised hardware and firmware integrity, vulnerabilities in network and high-speed interconnect technologies, improper multi-tenant isolation and resource reuse, insufficient security of out-of-band management planes, and compromise of the AI infrastructure supply chain; the remaining four are high risk and one is medium risk. OWASP notes that firmware malware may persist after reinstalling the system or wiping disk data, and that traffic over high-speed interconnects such as InfiniBand and RoCE may be difficult for traditional network security tools to fully monitor; enterprises choosing GPU cloud services should confirm whether the provider's security certification covers the underlying GPU infrastructure.