DeepSeek: Legal Considerations for Enterprise Users

DeepSeek, a Chinese artificial intelligence (“AI”) startup, recently made waves across the global AI landscape with the release of its latest open-source R1 model. While some versions of DeepSeek’s model can be downloaded and run locally without storing data on cloud storage controlled by DeepSeek,¹ enterprise users exploring use of DeepSeek through its iOS or Android app interfaces or its web chatbot interfaces must carefully consider the sweeping terms that govern the confidentiality, ownership, privacy, and security of data that is shared or generated through these DeepSeek interfaces. In this article, we discuss some of the key legal considerations.

On January 27, 2025, DeepSeek roiled U.S. technology stocks² when reports emerged that its R1 generative AI reasoning model, purportedly developed and trained at a fraction of the cost of OpenAI and Meta’s comparable models, had topped downloads in Apple’s App Store. The performance of DeepSeek’s AI model, which is open-sourced under an MIT License, is reportedly on par with OpenAI’s o1-mini model released in September 2024. However, DeepSeek reported that it achieved these performance levels with just about 5% of the development costs of its rivals. The revelation, while potentially a positive development for the evolution and adoption of generative AI models, has cast doubt on the popular narrative that advancement of large language models (“LLMs”) that power generative AI applications requires enormous capital and computing power. This has huge implications for investments in AI and valuations for startups focused on developing AI models, which have so far relied on far higher assumptions on costs.³

Some enterprise users may be considering swapping out their existing generative AI models for one of the several versions of DeepSeek that are now, or will likely soon become, available to realize potential savings. But doing so is not without risks. Because DeepSeek’s model is open-sourced, any user can download the model and run a local instance that would not require sharing data with DeepSeek or agreeing to DeepSeek’s terms governing its online platform interface versions, such as Android and iOS versions. Indeed, according to Pitchbook, there is already a surge of AI developers testing the DeepSeek model as an alternative to existing models from OpenAI.⁴ However, as DeepSeek does not currently offer an enterprise version of its online model, enterprise users who are considering running the online version rather than hosting their own local instances would be subject to DeepSeek’s standard version and its related terms of use. We highlight some of the key terms and associated legal considerations below.

1. DeepSeek Reserves for Itself Broad Rights in Personal Data. DeepSeek’s privacy policy generally gives DeepSeek broad rights to exploit user data that it collects either through prompts or from user devices, including “by monitoring interactions and usage across your devices, analyzing how people are using it, and by training and improving our technology” and to “comply with [DeepSeek’s] legal obligations, or as necessary to perform tasks in the public interest, or to protect the vital interests of [its] users and other people.”⁵ The privacy policy also gives DeepSeek the right to share information collected with its advertising and analytics partners, as well as to third parties in connection with a corporate transaction, such as a merger, sale of assets or shares, reorganization, financing, change of control, or other acquisition.⁶ All personal data is stored on servers in China.⁷ Companies should carefully review such privacy terms to ensure that no such provisions conflict with such companies’ own internal policies regarding data security and retention or external commitments to their customers or users. There are also trade laws that restrict or prohibit data transfers to certain foreign countries, including China, that may be implicated by the use of DeepSeek’s online platforms.
2. DeepSeek Retains Broad Rights to Use Inputs and Outputs, Including Confidential Information. DeepSeek grants users ownership rights in outputs generated by the services,⁸ while DeepSeek software repositories (e.g., V3) rely on the open-source MIT License for distribution.⁹ However, DeepSeek reserves the right to use inputs and outputs to “provide, maintain, operate, develop or improve” its services or the underlying technologies supporting the services.¹⁰ Given that DeepSeek could potentially reuse inputs or outputs for other users, users should not have any expectation that they will maintain exclusive control over their inputs or outputs in connection with the DeepSeek online platform. In addition, any sensitive or confidential information provided to DeepSeek’s online platform may be used to train the model, leading to possible public disclosure. It is important for enterprise users to establish clear policies and technical guardrails designed to prevent leakage of confidential or sensitive information through online services, including AI.
3. Users Bear Infringement and Other Liabilities and Risks. Any use of the DeepSeek platform is at users’ own risk. DeepSeek broadly disclaims any liabilities in connection with its services.¹¹ Little information is publicly available regarding the training data used for DeepSeek’s AI platform, or the guardrails or other “responsible AI” practices (if any) implemented by the platform. While other AI providers are increasingly willing to offer limited indemnification for paid subscription models (such as if certain output infringes third-party intellectual property rights)¹², DeepSeek does not indemnify users in any circumstance. Instead, users agree to indemnify, defend, and hold DeepSeek and its affiliates and licensors harmless against any liabilities, damages, and costs payable to third parties arising out of the users’ breach of the terms of use, their breach of applicable laws and regulations or a third party’s rights, their fraud or other illegal acts, or their intentional misconduct or gross negligence.¹³ It is not yet known whether any versions of DeepSeek’s online platforms will include technological filters to detect and avoid infringement of third-party copyrighted material, which could leave users exposed to such risks.
4. Users Must Comply with Attribution and Other Vague Requirements. Users must clearly indicate that the output was generated by AI, and must also proactively verify the authenticity and accuracy of the output.¹⁴ In addition, users must comply with other vague guidelines, such as using the services in adherence with the “principles of voluntariness, equality, fairness, and good faith” and not using the services to “seek unjust benefits” or “disrupt the normal order of the internet platform.”¹⁵ Users are also restricted from using the services for “dangerous purposes that may have serious harmful impacts on physical health, psychology, society, or the economy, or violate scientific and technological ethics” or “other uses… that may harm DeepSeek’s interests.”¹⁶ Interestingly, the terms of use also prohibit use of the “algorithms, data, platforms, etc. to implement monopolistic and unfair competition behaviors,” which could potentially be read to prohibit benchmarking if such practice is deemed anti-competitive.¹⁷ Users should expect DeepSeek to broadly interpret these clauses to limit access or use of the platform, or to otherwise seek claims of breach of the terms of use.
5. PRC Law and Venue Govern. DeepSeek’s terms of use are governed by the laws of the mainland of the People’s Republic of China.¹⁸ In the event of any dispute arising from the signing, performance, or interpretation of the terms of use, the parties must first attempt to resolve the dispute amicably, and if such negotiations fail, then either party has the right to file a lawsuit with a court having jurisdiction over the location of the registered office of Hangzhou DeepSeek.¹⁹ Foreign companies may not be familiar with litigating in China, and may not have the resources to pursue litigation in Chinese courts. If DeepSeek does not comply with its terms of use, a user would not be permitted to conduct pre-trial discovery for purposes of proving claims, which would essentially render DeepSeek “judgment proof” and leave companies practically without any legal recourse.

Conclusion

As with other AI models, it is critical that users carefully review DeepSeek’s terms of service (including licenses on platforms such as GitHub), privacy policy, and other user agreements to understand the legal risks that come with using its AI tools. As discussed above, DeepSeek’s terms of use grant it broad rights to use personal information, while shifting legal risks to users. While DeepSeek’s models offer exciting opportunities for the advancement of AI, users, particularly enterprise users, should ensure that their use of these models does not present an unacceptable level of technical, legal, or business risk.