(Yicai) April 7 -- China’s Asset Management Association has released the first technical standards governing the application of artificial intelligence models in the fund management industry to improve the security and compliance of AI use as adoption accelerates across the financial sector.

The guidelines divide AI applications in asset management into six layers -- infrastructure, data management, model services, application technology, scenario applications, and security management -- and set out specific technical requirements and management measures for each, according to the recent document released by the association.

The drafting institutions include nine major Chinese fund managers such as E Fund Management and China Asset Management, as well as AI technology providers including Alibaba Cloud and Huawei Technologies, the document showed.

The standards state that fund management institutions must not directly use non-anonymized customer personal information, core trading instructions, or undisclosed research data to train or fine-tune AI models. Training or inference involving highly sensitive investment research strategies or risk-control data is only permitted in dedicated model environments that are privately deployed and physically or logically isolated.

The guidelines also require institutions to establish model security mechanisms, including content security reviews, adversarial example protection, supply chain security assessments, and access control isolation, to ensure the compliance, robustness, and interpretability of model outputs.

In addition, institutions are required to adopt comprehensive risk-prevention measures covering multi-factor authentication, application programming interface protection, compliance audit traceability, and emergency response and recovery systems.

Since March, AI agent products represented by OpenClaw have gained popularity in China. However, such tools typically require elevated system permissions during operation, raising regulatory concerns about potential security risks. Several authorities, including the industry and information technology ministry and the National Computer Network Emergency Response Technical Team/Coordination Center of China, have issued risk alerts on AI agents.

Because fund management institutions operate closed trading systems, tools such as OpenClaw have limited relevance to their daily trading activities, a fund manager previously told Yicai. However, the use of AI for large-scale data processing and basic, repetitive tasks remains an important development trend in the industry, he added.

Editors: Dou Shicong, Emmi Laine