ETSI has released TS 104 223, a new specification establishing 13 core and 72 trackable cybersecurity principles to secure AI models and systems across five lifecycle phases. This baseline guidance addresses unique AI risks such as data poisoning, model obfuscation, and prompt injection by integrating established security practices with novel approaches tailored for the AI supply chain. Developed by the ETSI Technical Committee on Securing Artificial Intelligence, the standard aims to protect AI systems from evolving threats throughout their design, deployment, and maintenance. To support practical adoption, ETSI will also publish an implementation guide for small and medium enterprises featuring case studies across various deployment environments.
Keywords: AI security, lifecycle protection, data poisoning, model obfuscation, baseline requirements