DeepSeek May Deep Sink Your Innovation

Yesterday, I visited the DeepSeek site, and the first thing that caught my attention was the welcoming tone of its homepage. The screen greeted me with: “Hi, I’m DeepSeek. How can I help you today?” It was a warm and encouraging start to my DeepSeek experience. But it got me thinking—are we trading our data to China for the next large language model (LLM)?

An article published by Stanford sheds light on the staggering extent of government involvement in Chinese businesses. According to the article: “By conventional measures, China has 391,000 state-owned enterprises (SOEs), but new analysis of state ownership among all 40 million registered firms in China finds that 363,000 firms are 100% state-owned, 629,000 firms are 30% state-owned, and nearly 867,000 firms have at least some state ownership.” Interestingly, DeepSeek stands out as a rare case of private ownership in China, with no apparent government stake. On the surface, this might seem like a win for data protection. However, since DeepSeek is a Chinese company that trains its data on servers located in China, certain risks still arise.

For example, their terms of service state:

“User Input. When you use our Services, we may collect your text or audio input, prompt, uploaded files, feedback, chat history, or other content that you provide to our model and Services.”

I’m not a lawyer, but this statement seems straightforward. Anything you share with DeepSeek—text, files, audio, or images—can be used by the company. This is similar to ChatGPT’s end-user license agreement (EULA) for its free-tier service. However, even if DeepSeek is privately held, it operates in a jurisdiction where the Chinese government has the legal authority to access any data processed by Chinese companies.

There are two significant risks when using DeepSeek:

Data Exposure During Model Training

Your data—including prompts, audio, images, and uploaded files—could be exposed during the training process. A similar incident occurred in late 2023 when Samsung employees used the public version of ChatGPT with proprietary company data. ChatGPT’s EULA explicitly states that user input may be used for training purposes. Despite this warning, the excitement around AI led many to input sensitive information, ultimately resulting in Samsung’s intellectual property being leaked. Like ChatGPT, DeepSeek is eager for training data, meaning your input could also be inadvertently exposed.

Theft of Intellectual Property (IP)

This second scenario is even more concerning. China has a well-documented history of intellectual property theft, ranging from high-speed rail designs to telecommunications equipment. There are countless examples of Chinese entities appropriating IP from international companies. In 2023, Mike Burgess, the director-general of Australia’s Security Intelligence Organization, remarked:

“The Chinese government is engaged in the most sustained, scaled, and sophisticated theft of intellectual property and expertise in human history.”

Furthermore, Chinese contracts often include strict IP clauses favoring local companies, making it difficult for foreign businesses to pursue legal action when their technology is copied. If you’re using DeepSeek for innovation, you might unintentionally be handing over your intellectual property.

That said, it’s not all bad news. If you want to use the DeepSeek-R1 model, it’s available on Amazon for cloud-based access. Alternatively, you can run it locally on your own computer using Ollama, an open-source tool for deploying large language models.

_____________________________

Matt is a Director of Product Management for a leading mobile platform enablement company. He has traveled extensively in the United States and overseas for business and travel. His travels include India, Mexico, Europe, and Japan where he was an active blogger immediately following the Kaimashi quake. Matt enjoys spending time outdoors and capturing the world through the lens of his Nikon D90. He enjoys researching the political, economic, and historical influences of the places he visits in the world, and he commonly blogs about these experiences. Matt received a Bachelor in Computer Science at Mercer University, and is a noted speaker on innovation, holding over 150 patents. His remaining time is spent with his family going from soccer game to soccer game on the weekends.

As the Voice of the Veteran Community, The Havok Journal seeks to publish a variety of perspectives on a number of sensitive subjects. Unless specifically noted otherwise, nothing we publish is an official point of view of The Havok Journal or any part of the U.S. government.