NEWS  /  Analysis

Chinese AI Pioneer DeepSeek Labeled 'Profound Threat' to US National Security

By  xinyue  Apr 20, 2025, 10:24 p.m. ET

"DeepSeek poses a profound threat to US security. While it appears to be an ordinary AI chatbot, closer examination shows it funnels data to the People's Republic of China (PRC), exposes users to security risks, and runs on a model that secretly censors and manipulates content in line with Chinese law," the report said.

CFP

CFP

AsianFin -- Chinese AI company DeepSeek has been labeled a "profound threat" to US national security by a congressional committee, which has urged chipmaker Nvidia to disclose details about its customers and transactions with the start-up.

In a report released Wednesday, DeepSeek was described as being tightly linked to the Communist Party and "engineered to unlawfully erode US technological leadership and key national security policies."

The report also pointed to DeepSeek's connections to military research and strategic institutions, including Zhejiang Lab, a major contributor to China's scientific and technological advancements.

"DeepSeek poses a profound threat to US security. While it appears to be an ordinary AI chatbot, closer examination shows it funnels data to the People's Republic of China (PRC), exposes users to security risks, and runs on a model that secretly censors and manipulates content in line with Chinese law," the report said.

It cited research by Feroot Security, which alleged that DeepSeek transmits American user data to China through "back-end infrastructure linked to China Mobile" and incorporates "tracking tools" from Chinese tech firms like ByteDance and Tencent.

The committee labeled state-owned China Mobile as "military-related" and ByteDance as "controlled by a foreign adversary," while the Pentagon has classified Tencent as a Chinese military company.

"With its direct links to China's security and surveillance systems and its unchecked data gathering, [DeepSeek] could serve as an open-source intelligence asset channeling American user data into a hostile network," the House Select Committee on the Strategic Competition between the United States and the Chinese Communist Party reported.

"The threat is evident: millions of Americans are engaging with an AI tool designed to serve the CCP. Beijing is no longer limiting its internet controls to domestic users. It is embedding its Great Firewall into platforms used daily in the US."

China's tech ambitions have faced increasing scrutiny as US concerns over national security risks intensify.

Chinese embassy spokesperson Liu Pengyu rejected the claims as "groundless," telling Bloomberg the Chinese government respects data privacy and "has never required, nor will it require, companies or individuals to illegally collect or store data."

Earlier this week, Nvidia warned it may incur writedowns of up to US$5.5 billion this quarter due to new Trump administration restrictions on sales of its H20 chip—a lower-powered version of its top chips that complies with US export rules.

According to research firm SemiAnalysis, DeepSeek possesses over 60,000 of Nvidia's advanced chips, some of which now require licenses for export to China, to run its models.

Committee chairman John Moolenaar, a Michigan Republican, and the top Democrat on the panel, Raja Krishnamoorthi, jointly wrote to Nvidia CEO Jensen Huang, expressing concerns that DeepSeek could be skirting US export controls via third countries, citing recent chip fraud arrests in Singapore.

They requested Nvidia provide detailed records, including customer lists from Southeast Asia who bought over 499 AI-related chips since 2020, along with any communications with DeepSeek and contracts involving Chinese entities restricted under US law, with a response deadline of April 30.

Nvidia said Wednesday it strictly follows US export regulations, stating: "Our reported Singapore revenue reflects billing addresses, often for US customer subsidiaries. The products themselves ship to locations like the US and Taiwan, not China."

The report also revealed that US firm OpenAI had accused DeepSeek of training its language models using "unlawful" methods.

OpenAI testified that DeepSeek's R1 resembled its own models and accused DeepSeek staff of "bypassing safeguards" to access its intellectual property and reduce training costs through a method called distillation.

Experts say distillation is common and difficult to detect, though DeepSeek has not responded to the allegation.

The committee made several recommendations to guard against future "strategic surprises" in AI, including tighter export controls on AI and semiconductor tech and a federal ban on procuring Chinese AI systems.

The US's aggressive AI policy has drawn warnings that limiting exports of even lower-end chips might actually accelerate China's semiconductor development.

Just last week, Chinese telecom giant Huawei unveiled a "nuclear-level" AI infrastructure said to compete with Nvidia's offerings in resolving computing capacity challenges.

Please sign in and then enter your comment