ETSI has published report GR SAI 006, which outlines the specific hardware security requirements needed to protect AI and machine learning systems from attacks that exploit neural processors and specialized computing units. The document identifies common vulnerabilities in AI hardware and details mitigation strategies to ensure the integrity of both the AI systems themselves and the underlying infrastructure they rely on. Rapporteur Alec Brusilovsky emphasizes that achieving a high level of security for AI ecosystems depends fundamentally on robust hardware security measures. The report also reviews academic and industrial experiences regarding the use of AI to enhance hardware protection and provides a framework for addressing these emerging threats.
Keywords: AI hardware security, neural processors, hardware vulnerabilities, AI/ML attacks, neural networks