China Issues First National Standard on Generative AI Service Security

On February 29, the National Information Security Standardization Technical Committee released the "Basic Requirements for Generative AI Service Security," the first comprehensive national technical document dedicated to generative AI safety and a support to the interim measures issued in July of the previous year.

The Basic Requirements define fundamental safety criteria for generative AI services, including corpus safety, model safety, and specific security measures, and they provide detailed security assessment guidelines. They apply to providers delivering generative AI services to the public within China, whether conducting self‑assessments, commissioning third‑party evaluations, or assisting regulatory bodies in judging service safety levels.

Ant Group, as one of the primary contributing units, deeply participated in the discussion and drafting of national standards such as "Generative AI Security Basic Requirements," "Safety of Pre‑training and Optimized Training Data," and "Data Annotation Security." The company also translated its large‑model security testing experience into the Cloud Security Alliance (CSA) international standards.

While the emergence capabilities of large models inject strong momentum into the digital world, they also bring complex, uncontrollable risks such as AI hallucinations, data leakage, unpredictable outputs, security threats, discrimination, and bias.

Ant Group places great emphasis on AI safety and technology ethics, applying the technical document requirements to its own practice through its integrated large‑model security solution "Ant Tianjian," continuously ensuring that generative AI remains safe, trustworthy, and controllable. The company looks forward to co‑building with the industry to address new challenges posed by AI development and to provide solid security guarantees for the digital economy.