DeepSeek: Threat to US National Security
House Committee Report: Deepseek National Security Threat
A bipartisan report from the House Select Committee has designated the Chinese startup as aartificial intelligence DeepSeek Ltd. as a “profound threat” to U.S. national security. The commission accuses the company of serving as an “open-source intelligence asset” by redirecting data on U.S. users to China.
DeepSeek, a laboratory ofIA China-based company gained notoriety earlier this year with the launch of its open-source language model, DeepSeek-R1, which is notable for its optimization for reasoning. This model outperforms some of the most advanced models from companies such as OpenAI and Anthropic on a range of complex tasks, and does so at a much lower training cost.
Shortly after DeepSeek R1's launch, OpenAI and its main investor, Microsoft Corp., denounced fraudulent practices, claiming the model was trained using stolen data from their own AI models. They accused DeepSeek of using distillation techniques, which involve studying the output of other models and replicating them to speed development.
The report cites testimony from OpenAI, which claims that DeepSeek “circumvented” its models’ protections to extract their reasoning outputs, thereby accelerating R1’s development. OpenAI also claims that DeepSeek likely used open-source AI models to create synthetic data for training R1.
Ironically, OpenAI has been criticized for harvesting content from the internet to train its own models, without asking permission from content creators.
In addition to the OpenAI data theft, the commission accuses DeepSeek of “redirecting US user data” to China through unsecured networks, noting that this information could be an intelligence asset for the Chinese Communist Party. However, it is essential to distinguish between the R1 model and DeepSeek’s consumer-facing chat application, which is similar to OpenAI’s ChatGPT. This service is hosted on servers in China, which are likely collecting user data. In contrast, R1 is an open-source model that can be hosted on any server around the world, including those of US companies like Microsoft and Meta Platforms Inc.
The commission also warned that DeepSeek R1 censors its responses to suppress information related to democracy, Taiwan, and Hong Kong, without informing users. Furthermore, the report notes that DeepSeek founder Lian Wenfang has close ties to the Chinese Communist Party and is “ideologically aligned with Xi Jinping Thought.”
While the White House has yet to comment on the commission's report, it appears to be taking action against DeepSeek and other Chinese AI companies. Last month, reports surfaced that it might ban government employees from using the R1 model. Earlier this week, Nvidia revealed that the government had imposed new restrictions on its ability to ship its powerful graphics processors, widely used for training AI models, to China.
Nvidia has been instructed to secure a special export license to ship its H20 processors to China in the future. However, the difficulty of obtaining this license effectively amounts to a ban on exporting these chips. Although the H20 processor was developed specifically for Chinese companies in response to previous restrictions, DeepSeek appears to have used it to train the R1 and other models.
It remains to be seen whether tighter restrictions will hamper DeepSeek's progress. Critics point out that previous restrictions forced the company to develop more efficient models on lower-performance processors, warning that this will intensify China's efforts to develop its own chips.
