Published on 9. October 2026
Reading time approx. 2 Minutes

AI Compliance in China: New NFRA Guidelines for the Financial Sector

  • The NFRA has published the first comprehensive AI guidelines for financial institutions
  • Banks and insurers bear full responsibility for AI applications
  • Risk management and human oversight are made mandatory
  • The guidelines set new standards for AI compliance in China
Sebastian Wiendieck
Partner
Attorney at Law (Germany)
Li Wang, LL.M.
Associate Partner
Legal Counsel (China)
With the NFRA guidelines, China has, for the first time, established a comprehensive regulatory framework for the safe use of AI in the banking and insurance sectors. The requirements range from governance structures and data management to risk controls for high-risk applications. Financial institutions should review and adapt their existing AI processes without delay.

Sector-specific regulations: focus on the financial sector

Regulatory background

The opinion of the Supreme People’s Court is not the first regulatory framework for AI in China. Even immediately prior to the publication of the opinion, specific regulations had already been enacted in individual sectors. For the financial sector, the “Guidelines on the Safe Development and Application of Artificial Intelligence in the Banking and Insurance Sectors” of 18 June 2026 (Jin Fa [2026] No. 8) are particularly relevant.

The NFRA guidelines are China’s first systematic regulatory document specifically aimed at the security of AI applications in the banking and insurance sectors. They comprise 32 guidelines across seven regulatory areas: governance architecture, development and application, data governance, computing power, risk management, capacity building, and supervision and enforcement.

Contrary to the opinion of the Supreme People’s Court, the NFRA Guidelines constitute binding regulatory requirements for financial institutions. They are therefore directly relevant to compliance for banks and insurance companies.

Key points of particular relevance to financial institutions

The guidelines establish the principle that financial institutions, as users of AI technology, bear full responsibility. They must clearly define the division of tasks and responsibilities between developers, providers and users, and set out the responsibilities at every stage of the internal process chain.

Financial institutions must establish a management system covering the entire lifecycle of AI applications: from needs analysis through data preparation, training, deployment and maintenance to evaluation and decommissioning. In doing so, they must carry out a risk-based classification and categorisation of AI applications.

Human oversight and governance

For high-risk applications, particularly in areas such as credit approval, claims settlement, asset valuation and risk management, financial institutions must establish human oversight and intervention mechanisms at key stages of the process. The guidelines emphasise human-machine collaboration and the clear definition of functional boundaries, as well as system and data access rights.

Data protection and model control

The guidelines expressly prohibit the use of personal information (names, ID numbers, mobile phone numbers, bank card numbers, etc.) for the training and optimisation of generative AI models, in order to prevent data breaches. Generative AI models sourced externally must also be registered with the Cyberspace Administration.

Conclusion

The NFRA guidelines make it clear that AI governance in the financial sector is no longer a voluntary best practice, but rather a requirement for regulatory compliance. At the same time, they serve as an example of the organisational requirements that are likely to become increasingly important in other regulated sectors in future