ETSI publishes security standard for AI computing platforms
A new ETSI technical specification establishes security requirements for the computing platforms that support AI model training and inference, with a focus on protecting models, data, and platform infrastructure.
European Telecommunications Standards Institute has published ETSI TS 104 033, a new technical specification defining security requirements for AI computing platforms.
The specification focuses on the infrastructure layer that hosts AI model training and inference. According to ETSI, these platforms provide the execution environment and resources needed to run AI systems and therefore play a central role in securing the wider AI ecosystem.
The standard establishes a baseline set of security requirements and functions designed to help operators mitigate threats affecting AI platforms and the assets they process, including models and datasets.
Among its key elements, the specification defines security components and service interfaces that can be used to implement the required protections. It also addresses protection of AI models and data in different states, including when data is stored, transmitted, or actively being used.
ETSI said the specification is intended to help reduce risks such as model extraction, unauthorised access, and data leakage.
The document also aligns with the principles set out in ETSI EN 304 223, a framework covering the lifecycle of AI systems from design and development through deployment, maintenance, and decommissioning.
