en.Wedoany.com Reported - On June 2, the European Telecommunications Standards Institute (ETSI) released Technical Specification TS 104 033 in Sophia Antipolis, France, defining security requirements for AI computing platforms. This specification focuses on the platform layer that hosts AI model training and inference, aiming to establish a clearer security baseline for data centers, edge computing, and enterprise AI deployments.
The security risks of AI systems are expanding from the models themselves to the operating environment. Model training, inference services, dataset calls, parameter updates, interface access, and computing resource scheduling all rely on the underlying computing platform. If the platform lacks unified security components, service interfaces, and asset protection mechanisms, model extraction, data leakage, unauthorized access, runtime tampering, and supply chain attacks can infiltrate AI systems through the platform layer. With the release of TS 104 033, ETSI focuses not on discussing AI application scenarios, but on shifting security controls forward to the computing infrastructure that hosts model operations, enabling platform designers, system integrators, and operators to architect around mandatory security requirements.
The specification covers security requirements and functions, security components and service interfaces, as well as protection requirements for AI models and data in use, in transit, and at rest.
From an industrial application perspective, AI computing platforms have entered data centers, edge nodes, industrial sites, and government-enterprise private environments. When deploying generative AI, industry-specific large models, and intelligent agent systems, enterprises often simultaneously use their own computing power, cloud platforms, open-source models, third-party tools, and internal business data. Without standardized references for platform security, it is difficult for enterprises to determine which capabilities are mandatory, which interfaces need isolation, and which model assets require prioritized protection. By defining the security requirements that computing platforms must meet, TS 104 033 provides a more unified evaluation framework for AI infrastructure design and also helps suppliers incorporate "security by default" as part of platform capabilities during product development.
This specification also aligns with ETSI's previously released EN 304 223. EN 304 223 establishes baseline cybersecurity requirements for AI models and systems, covering lifecycle stages such as secure design, development, deployment, maintenance, and decommissioning; TS 104 033, on the other hand, focuses on the AI computing platform itself, supplementing security controls for the model operating environment. As regulatory requirements such as the EU AI Act gradually take effect, developers, platform providers, data managers, and operators within the AI system supply chain will need clearer technical evidence to demonstrate that their systems possess governable, auditable, and maintainable security capabilities.
Subsequent impacts will depend on adoption by cloud service providers, AI infrastructure companies, edge computing equipment vendors, and industry users. For the AI industry, security standards are no longer just compliance documents; they will influence platform procurement, system integration, model hosting, and cross-border business deployments. ETSI's release of TS 104 033 signifies that competition in AI infrastructure is extending beyond computing scale, model performance, and deployment costs to encompass platform security, data protection, and supply chain trustworthiness.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
