TC260-005 — Ethics-Safety Guidelines for AI Applications 1.0
Status
Section titled “Status”TC260-005, Ethics-Safety Guidelines for Artificial Intelligence Applications 1.0, was published on 19 May 2026 by China’s National Information Security Standardisation Technical Committee. It is a principle-based reference technical document, not a mandatory national standard.
Function
Section titled “Function”The guidelines provide ethical and safety principles, baseline requirements and lifecycle guidance for parties developing, supplying, deploying and using AI applications. They are designed to operate together with binding rules on personal information, automated decision-making, content labelling, algorithm governance and intellectual property.
Its publication extends the TC260 layer beyond service-security and filing requirements into broader application ethics and sociotechnical risk governance, including generative AI and agents.
Structural impacts highlighted by TC260
Section titled “Structural impacts highlighted by TC260”TC260’s accompanying expert interpretation says the Guidelines organise risk around six structural impacts. Its published commentary expressly highlights three combined groups:
- erosion of human control and pressure on environmental sustainability;
- disruption of basic social order, including structural inequality, bias and discrimination; and
- distortion of individual cognition and infringement of personal or organisational rights.
These are analytical risk categories rather than prohibited-use clauses. The Guidelines use them to organise governance across research and development, design and manufacture, deployment and use.
Compliance significance
Section titled “Compliance significance”- Non-binding status: TC260-005 is guidance and does not itself create an administrative penalty or market-access gate.
- Interpretive value: it supplies a common ethics-and-safety vocabulary for internal reviews, procurement, product assessment and future standards.
- Relationship to hard law: PIPL, the algorithm and generative-AI rules, content-labelling duties and sector-specific law continue to determine binding obligations.
- Agent relevance: its emphasis on human control, cognitive dependency, rights protection and lifecycle accountability complements the May 2026 AI-agent Opinion and the binding Anthropomorphic Interaction Measures.