CNIPA Guidelines on AI-Related Inventions: What Foreign Applicants Need to Know

Introduction

On December 31, 2024, the China National Intellectual Property Administration issued the Guidelines for Patent Applications for AI-Related Inventions (Trial).

Further amendments to the Patent Examination Guidelines took effect on January 1, 2026. Together, these rules clarify the examination of inventions involving artificial intelligence algorithms, models, big data and AI applications.

AI-related inventions remain subject to the general requirements of Chinese patent law, including patent eligibility, novelty, inventive step, practical applicability and sufficient disclosure.

1. Inventorship

An AI system cannot be named as an inventor.

Only natural persons who have made a creative contribution to the substantive features of an invention may be named as inventors. Companies, research teams, software platforms and AI systems do not qualify.

The use of AI tools during research and development does not prevent an invention from receiving patent protection, provided that one or more natural persons have made the required creative contribution.

Applicants should therefore identify the relevant human inventors accurately and retain internal records of their contributions.

2. Technical Character and Patent Eligibility

AI-related claims are assessed as a whole.

An invention is more likely to qualify as patent-eligible subject matter where its algorithmic features interact with technical features to solve a technical problem and produce a technical effect.

For example, an AI model may form part of a patentable technical solution where it is applied to image processing, industrial control, medical equipment, manufacturing, communications or another technical field.

By contrast, claims directed only to abstract algorithms, mathematical rules, business rules or mental activities may not be eligible for patent protection.

Applicants should clearly explain:

• The technical problem addressed by the invention;

• The technical features used to solve that problem;

• How the algorithm or model interacts with those features; and

• The technical effect produced by the claimed solution.

3. Inventive Step

The use of artificial intelligence does not automatically make an invention inventive.

Simply applying a known model to a different object or business scenario may be considered obvious if no meaningful technical adaptation has been made.

Algorithmic features may support inventive step where they are specifically adapted to a technical application and work together with technical features to produce an improved technical effect.

Applicants should clearly describe any changes made to:

• The model structure;

• The training process;

• Data-processing methods;

• Input or output configurations; and

• The interaction between the model and the technical environment.

Where appropriate, comparative data or testing results may help demonstrate the claimed technical improvement.

4. Sufficiency of Disclosure

The specification must contain enough information to enable a person skilled in the relevant field to implement the invention.

For inventions involving the construction or training of an AI model, the application should disclose the necessary model structure, training steps, relevant parameters and technical relationships.

For inventions applying an existing model to a specific technical field, the specification should explain:

• How the model is integrated into the technical scenario;

• How input and output data are selected and configured;

• The relationship between the data, model processing and technical result; and

• How the claimed technical effect is achieved.

Applicants are not required to disclose every implementation detail, but broad functional claims must be adequately supported by the specification.

5. Legal and Ethical Compliance

AI-related applications may be refused where data collection, rule setting, model decisions or other features violate Chinese law, social morality or the public interest.

Applicants should carefully review the use of:

• Personal information;

• Facial-recognition data;

• Medical or health data;

• Training datasets;

• Automated recommendation systems; and

• Algorithmic decision-making mechanisms.

The specification should avoid presenting technical solutions that rely on unlawful data collection, discriminatory decision-making or other legally or ethically problematic practices.

Practical Implications for Foreign Applicants

Foreign applicants filing AI-related patent applications in China should:

• Identify the human inventors accurately;

• Define the technical problem clearly;

• Explain how the AI model contributes to the technical solution;

• Provide sufficient implementation details;

• Support claimed technical effects where appropriate;

• Review data and privacy compliance before filing; and

• Ensure that the claims are adequately supported by the specification.

A pre-filing review by experienced Chinese patent counsel can help identify issues relating to patent eligibility, inventorship, disclosure and data compliance.

Strategic Recommendations

Applicants should consider the following steps:

1. Maintain internal records of the human inventors’ creative contributions.

2. Avoid describing the invention solely as an algorithm, model or business process.

3. Explain the technical relationship between the AI features and the practical application.

4. Include appropriate model, training and data-processing information.

5. Use comparative results or experimental data where necessary to support technical effects.

6. Review whether data collection, model decisions and other features comply with Chinese law, social morality and the public interest.

7. Ensure that the scope of the claims does not exceed the disclosure provided in the specification.

For assistance with drafting, reviewing or prosecuting AI-related patent applications in China, contact the Reehor IP team.