ETSI has released its first report addressing the critical need for standards to secure AI-based automated networks, moving beyond ethical discussions to tackle technical vulnerabilities. The group defines artificial intelligence as systems capable of handling explicit and implicit representations to perform intelligent tasks, noting that current feasibility is driven by advancements in machine learning and deep learning. The report highlights specific security challenges such as adversarial learning, where training sets include disruptive samples, and real-world threats like deepfakes and malware obfuscation. To address these issues, ETSI is concurrently developing a threat ontology, guidelines for securing the AI data supply chain, and testing methodologies. This work aims to establish a comprehensive framework for protecting the increasingly complex landscape of AI technologies.
Keywords: AI security, threat ontology, adversarial learning, deepfakes, data supply chain