(Yicai) April 7 -- China’s fund industry has released the first standards on how artificial intelligence can be used in asset management, prioritizing security, compliance, and controlled deployment, as the technology’s adoption gathers pace across the financial sector.

The standards, recently issued by the Asset Management Association of China, divides AI use in asset management into layers -- infrastructure, data management, model services, application technology, scenario applications, and security management -- and assigns specific technical and governance requirements to each.

The standards were drawn up by the AMAC itself, nine major Chinese fund managers such as E Fund Management and China Asset Management, and AI technology providers such as Alibaba Cloud and Huawei Technologies.

They state that fund management institutions must not feed non-anonymized customer data, core trading instructions, and non-public research information directly into AI model training or fine-tuning. Training or inference involving highly sensitive investment strategies or risk-control data is only permitted in privately deployed environments that are physically or logically isolated.

The guidelines also require institutions to set up model security systems, including content reviews, adversarial sample defenses, supply chain security checks, and access segregation controls to keep outputs compliant, robust, and explainable.

They must also adopt comprehensive risk-prevention measures covering multi-factor authentication, application programming interface protection, compliance audit trails, and incident response and recovery systems.

The new standards come as AI agents such as OpenClaw have become popular in China, with regulators -- such as the Ministry of Industry and Information Technology and the National Computer Network Emergency Response Technical Team/Coordination Center of China -- warning that the elevated system permissions these tools need can create security weaknesses.

Because fund managers operate closed trading systems, tools like OpenClaw have limited relevance to their daily trading activities, one previously told Yicai. But the use of AI to process vast amounts of data and handle other basic, repetitive work remains an industry trend, he pointed out.

Editors: Dou Shicong, Emmi Laine