US House Select Committee Report Accuses DeepSeek of Spying and Circumventing Export Controls on Chips

Amalia Huot-Marchand is a graduate student at the Medill School of Journalism specializing in politics, policy, and foreign affairs.

On April 18, the United States House Select Committee on the Chinese Communist Party (CCP) released a report claiming the state-subsidized Chinese AI firm, DeepSeek, is a Chinese espionage tool and is using advanced Nvidia AI chips in violation of US export rules. The Select Committee also announced it would open an investigation into Nvidia’s chip sales in Asia.

This Select Committee’s steps appear to be aligned with the White House, which took a significant step to curb China’s growing AI developments. On April 15, the Trump administration told Nvidia it would require a license to sell its H20 chips to China. This move will seriously harm Nvidia’s profit this quarter by $5.5 billion, according to the company. The announcement came after Nvidia promised to invest $500 million to manufacture its chips entirely in the United States.

These moves underscore the growing intensity of the US-China rivalry in AI development. China is now only three months behind on the technology, an unnamed US AI industry executive told the Select Committee, according to its report. Both the Biden and Trump administrations imposed export controls to prevent China from exploiting US technology and to maintain American dominance in AI. The report, titled “DeepSeek Unmasked: Exposing the CCP’s Latest Tool for Spying, Stealing, and Subverting US Export Control Restrictions,” claimed multiple aspects of the AI chatbot could pose serious national security risks.

DeepSeek has been under scrutiny since its inception. In January, DeepSeek released a new model called DeepSeek R1. This model rivals the well-known AI models developed by firms such as OpenAI, Anthropic, and Google. Experts disagree on the extent to which the model was trained on fewer resources than its Silicon Valley rivals, but acknowledge that the efficiencies introduced by DeepSeek do advance the state of the art and are now likely to be adopted by other developers.

Behind DeepSeek’s rapid evolution is powerful American-made hardware. The report claimed that DeepSeek is using Nvidia’s advanced computer chips, also known as H100s, which are not permitted to be sold to Chinese buyers. In 2022, President Joe Biden imposed stricter export controls on advanced AI chips to prevent China from having access to the best AI technology in the market, expanding those restrictions at the end of his term.

In response to these regulations, Nvidia created less powerful chips, the H20. This allowed Nvidia to keep access to the Chinese market while selling less efficient chips in China. However, the Select Committee report claims that DeepSeek still had access to the H100. SemiAnalysis, an analytics firm, estimated that DeepSeek uses at least 60,000 Nvidia chips, including H100, H20, A100, and H800 chips. China can also easily buy H100 AI chips through third-party countries, like Singapore.

“We now know this tool exploited US AI models and reportedly used Nvidia chips that should never have ended up in CCP hands. That’s why we’re sending a letter to Nvidia to demand answers. American innovation should never be the engine of our adversaries’ ambitions,” said the Chairman of the Select Committee, John Moolenaar (R-MI), in a statement.

Nvidia responded in a statement, “The US government instructs American businesses on what they can sell and where – we follow the government’s directions to the letter. NVIDIA protects and enhances national security by creating US jobs and infrastructure, promoting US technology leadership, bringing millions of dollars of tax revenue to the US treasury, and alleviating the massive US trade deficit.”

In order to end China’s procurement of these chips, the Select Committee recommended an expansion of US export controls by increasing federal funding for the Department of Commerce’s Bureau of Industry. Furthermore, it proposed creating incentives for whistleblowers to denounce any infraction of export controls by providing whistleblower protection and a monetary incentive.

This competitive AI race may come at the expense of ethical development. “I don't think it's to anybody's benefit to be engaging in the Cold War on this because people figure out a way to do things differently, cheaper, and better,” said Julia Stoyanovich, Institute Associate Professor of Computer Science & Engineering at the Tandon School of Engineering to the House Committee on Space and Technology subcommittee hearing on April 8.

Experts suggested that focusing solely on the threats of DeepSeek and not its achievements might hinder the federal government from learning lessons from its Chinese counterparts.

“However, the extraordinary attention focused on DeepSeek is justified, even if the conclusions some have drawn from its success are not. It would be a great mistake for US policymakers to ignore DeepSeek or to suggest that its accomplishments are merely a combination of intellectual property theft and misleading Chinese propaganda,” said the Center for Strategic and International Studies in a report in March. “Policymakers need to understand that—even while DeepSeek has in some cases simply implemented innovations already known to US AI companies—DeepSeek has also demonstrated genuine technological breakthroughs of its own. These facts deserve careful consideration as the second Trump administration sets its AI policy agenda.”

Another issue exposed in the report is data privacy.

According to DeepSeek’s own policy, all user data is stored in mainland China on secure servers. However, DeepSeek is state-subsidized, meaning the data could be shared with state authorities, the Chinese Communist Party. Additionally, DeepSeek has ties to companies such as ByteDance, Tencent, and Baidu. The US government has flagged these companies for posing national security concerns, including surveillance, because they collect enormous amounts of data from American users.

“For these reasons, it is evident that the DeepSeek website and app act as a direct channel for foreign intelligence gathering on Americans’ private data,” stated the Select Committee's report.

Still, experts caution against framing China as the sole actor collecting user information. As NYU’s Stoyanovich said, “Our domestic models also ingest all of our data, and these are commercial models, and they don't tell us how they're using this data. So for us to be able to take the moral high ground and say that China shouldn't be doing things in a particular way, we have to start with our own environment here.”

DeepSeek is also subject to Chinese censorship laws. It excludes any information on various subjects the CCP sees as dangerous or controversial, including information about Taiwan, Hong Kong, and Uyghurs’ humanitarian rights. American DeepSeek users are subject to either misinformation or curtailed information that benefits China. The Select Committee report stated that when DeepSeek is asked about politically sensitive issues, 85% of its responses were censored or contained misinformation.

“Unlike AI models in open societies, DeepSeek exists in an ecosystem where compliance with state ideology is a prerequisite for survival,” said the report.

While the US and China are engaged in what some refer to as an AI cold war, this does not benefit overall research, nor does it protect everyday users. “I think the only way for us to win at this and to make sure that our national security is preserved, and our competitive advantage is preserved through openness. We need to make sure that we figure out how open models and open innovation can be supported in this country,” said Stoyanovich. “And open doesn't mean we give away all our secrets, and we give away all of our advantage. Openness means control because we set the rules for how data is shared, for how data is used, for what is disclosed, for whether we know how the systems were checked for safety, for correctness.”

Members of the Select Committee suspect DeepSeek of having “stolen” access to US AI models such as ChatGPT. This practice, known as model distillation, enables smaller AI tools to utilize outputs from larger AI tools, thereby reducing the cost of development and training. They accused DeepSeek personnel of purchasing accounts on false pretenses.

OpenAI told the committee that the “reasoning structures and phrase patterns” in DeepSeek outputs resemble OpenAI model outputs. The company asserted “high confidence” that DeepSeek broke OpenAI’s usage agreement rules prohibiting model distillation.

DeepSeek has not responded to requests for comment.