ETSI EN 304 223 establishes a comprehensive, lifecycle-based security framework specifically designed to protect AI systems against unique threats like data poisoning and model obfuscation. The standard defines 13 principles across five phases—secure design, development, deployment, maintenance, and end-of-life—to ensure AI models, including generative systems, are resilient and trustworthy by design. This collaborative effort aligns with international AI lifecycle models to provide a clear baseline for vendors, integrators, and operators while supporting the broader ecosystem. An upcoming technical report, ETSI TR 104 159, will further extend these principles with prescriptive specifications for generative AI domains such as deepfakes and misinformation.
Keywords: AI security, lifecycle-based requirements, data poisoning, model obfuscation, secure design