To meet legal and compliance requirements, DeepSeek reserves the right to review how users use the services through technical means, including risk filters and databases of illegal-content features. This is a reserved compliance power, not a statement that every interaction receives human review.
Supporting quotations
3.3 To fulfill legal and compliance requirements, DeepSeek has the right to use technical means to review the behavior of users using the Services, including but not limited to establishing risk filtering mechanisms, and creating databases for illegal content features.
Use & Processing · Context
Privacy and model details are in separate documents
DeepSeek directs users to its Privacy Policy for personal-information collection, protection and use, and to its model-mechanism document for training and model details. These links do not supply those documents’ contents in this agreement.
The linked Privacy Policy and model-mechanism document are not supplied here; their detailed practices and rules cannot be determined from these links.
Use & Processing · Company practice
Inputs produce model-generated Outputs
Inputs include text, images, files and other information. DeepSeek describes its models as computing and inferring responses, including text, tables and code; users can give feedback such as likes and dislikes. Outputs can contain errors, are for reference only and are not professional advice or DeepSeek’s views. Users remain responsible for decisions and consequences based on them.
Supporting quotations
1.2 The generative AI products and services provided by DeepSeek are based on large language models, which are built using neural network, developed through stages of large-scale self-supervised pre-training and targeted optimization training. These models can predict the next token by encoding and computing the input information (including text, images, files, and more), thereby possessing text generation and conversational abilities. They are adept at performing a wide range of text generation tasks and can be integrated into various downstream systems or applications. Specifically, within DeepSeek's product services, these models, based on user input information (referred to as "Inputs"), compute and infer to output corresponding content as a response (referred to as "Outputs" ), including text, tables, and code. Users can evaluate the output, including actions like liking or disliking, to provide feedback on their opinions about DeepSeek's output information.
5.1 Any and all Outputs provided by this Service are generated by the foundation model and shall not constitute DeepSeek's legal declaration of intent, nor shall the Outputs represent DeepSeek's views or opinions in any event. If any Output references any third-party products or services, this does not mean that DeepSeek endorses that Output. The Outputs may contain errors or omissions and are for your reference only. You should NOT treat any Outputs as professional advice. In particular, when using this Service to consult on medical, legal, financial, and other professional issues, please note that the Output does not constitute any advice or commitment, nor does it represent opinions in any professional field. If you require related professional services, you should consult professionals and make decisions under their guidance. The Output of our Services shall not form the basis for your further actions or omissions. You are solely and independently responsible for any decisions or actions you take based on any Outputs, as well as for any consequences that follow.
Use & Processing · User right or choice
You can opt out of the stated Input and Output reuse
DeepSeek may use Inputs and Outputs to a minimal extent to provide, maintain, operate, develop or improve the services or supporting technologies, on the stated conditions of secure encryption, strict de-identification and processing that cannot be reversed to identify specific people. You can opt out of this described processing by turning off “Improve the model for everyone.” This clause does not by itself describe all processing under the separate Privacy Policy.
Supporting quotations
4.3 Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by turning off "Improve the model for everyone".
Use & Processing · Context
Search may help accuracy but does not guarantee it
Turning on Search may improve accuracy to some extent. DeepSeek says AI inaccuracies cannot be completely avoided given current technology, so the feature does not remove the need to use the service rationally under the Terms.
Supporting quotations
5.2 When you enable the "Search" function, the accuracy of the Outputs may be improved to some extent. However, limited by current technical capabilities, inaccuracy in AI-generated content is a common industry-wide issue and cannot be entirely avoided. Therefore, please use our Services rationally in accordance with these Terms.
Use & Processing · Context
A separate document explains training and generation
DeepSeek links its Model Principles and Training Methods Explanation as a resource about training and content generation and guarding against misuse risks. The linked text is not supplied here, so this card does not describe its specific methods.
Supporting quotations
5.5 We have published the "Model Principles and Training Methods Explanation" to clarify the foundation model's training process and content generation mechanisms, helping you understand AI technology and guard against various risks that may arise from misuse or improper use.
Not stated here
The linked model explanation is not supplied; its specific training methods and mechanisms cannot be established here.