TC260 issues 25 national cybersecurity standard plans, including multiple AI safety projects
TC260下达25项国标计划，含AI安全项目
The TC260 Secretariat stated that the National Standardization Administration issued 25 recommended national cybersecurity standard plans, managed by TC260. This is a plan issuance notice and does not mean these standards have been completed or are in effect. AI safety-related plans in the attachment include artificial intelligence foundation model safety testing methods, on-device large model cybersecurity guidelines, artificial intelligence model development safety guidelines, artificial intelligence safety guardrail technical requirements, generative artificial intelligence agent system operation safety specifications, artificial intelligence training and inference framework safety requirements, and artificial intelligence user personal information protection guidelines. The attachment also lists plans related to agent data safety, model open-source safety, and large model safety specifications, as well as projects such as artificial intelligence safety testing institution capabilities and artificial intelligence-generated content detection, covering multiple safety topics including models, agents, data protection, and testing.
This batch of national standard plans covers safety requirements for foundation models, on-device large models, agents, and training and inference frameworks, which can be used to observe the layout direction of domestic AI safety standards.