A bipartisan U.S. House committee released a detailed report on April 16, casting Chinese AI firm DeepSeek as a serious national security risk involved in espionage, technology theft, and propaganda aligned with the Chinese Communist Party (CCP).
The Select Committee on the CCP, led by Chairman John Moolenaar (R-MI) and Ranking Member Raja Krishnamoorthi (D-IL), also revealed it is pursuing answers from U.S. chipmaker Nvidia regarding how its potentially export-controlled semiconductors allegedly fuel DeepSeek’s operations.
The committee’s investigative report, titled “DeepSeek Unmasked,” alleges the AI platform runs on hardware incorporating tens of thousands of Nvidia processors restricted from sale to the People’s Republic of China (PRC).
Citing estimates from analytics firm SemiAnalysis, the report suggests DeepSeek may possess over 60,000 Nvidia chips, including specific counts of A100, H100, H800, and the recently blocked H20 models. The committee posits these chips might have been acquired illicitly, potentially using nations like Singapore as transshipment points – a possibility underscored by a prior U.S. investigation.
Export Controls Tighten as Nvidia Faces Questions
This hardware focus lands amid stricter U.S. export controls. Effective April 15, 2025, the Department of Commerce effectively barred Nvidia H20 chip sales to China by mandating an “indefinite license,” citing national security worries. Commerce Department spokesman Benno Kass confirmed the move, stating the department acts “to acting on the president’s directive to safeguard our national and economic security.”
Consequently, Nvidia disclosed a US$5.5 billion financial charge tied to the now-unsellable H20 chips, which had been particularly valued for potentially superior AI inference performance compared to the already-banned H100.
In the ban’s immediate aftermath, Nvidia CEO Jensen Huang visited Beijing, meeting trade officials and, according to the Financial Times, DeepSeek founder Liang Wenfeng.
Responding later to the committee’s formal letter, Nvidia asserted it adheres to export rules “to the letter,” suggesting Singapore revenue figures often reflect billing addresses, not final chip destinations, and stating shipments go elsewhere but “not to China.”
The committee’s inquiry, however, seeks specific customer lists for large AI chip purchases in ASEAN/China and communications between Nvidia and DeepSeek. Meanwhile, Chinese domestic chip firms are responding; Huawei, for instance, is preparing mass shipments of its Ascend 910C AI chip, reportedly using advanced chiplet integration to target H100-level capabilities, starting in May 2025.
Data Harvesting, Propaganda, and IP Theft Alleged
The House report contends DeepSeek functions as a CCP intelligence tool, alleging it funnels extensive American user data – including chat history, device details, and “even the way a person types” – back to PRC servers.
This is reportedly channeled via backend infrastructure linked to China Mobile, a state-owned entity designated a Chinese Military Company by the US DoD.
An earlier investigation by Feroot Security, verified for the Associated Press by academic experts, identified code in DeepSeek’s login potentially connecting to China Mobile’s services. Expert Joel Reardon noted it was “clear that China Mobile is somehow involved in registering for DeepSeek.” The committee also states DeepSeek integrates tracking tools from other firms like ByteDance, Baidu, and Tencent known for surveillance roles.
Beyond data collection, the report asserts DeepSeek enforces CCP censorship, manipulating or suppressing responses on politically sensitive topics (like democracy or Taiwan) in 85% of cases without user disclosure.
DeepSeek’s origins are linked in the report to founder Liang Wenfeng, his quantitative trading firm High-Flyer Quant (which provided reported $420M funding and access to its 10,000 A100 GPU ‘Firefly’ supercomputer), the state-backed Hangzhou Chengxi tech corridor, the strategic Zhejiang Lab, and Liang’s academic advisor Xiang Zhiyu, whose research includes military applications.
The committee further alleges intellectual property theft, stating it’s “highly likely” DeepSeek used “unlawful model distillation techniques” to copy capabilities from leading U.S. AI models, specifically naming OpenAI. The report claims DeepSeek personnel “infiltrated U.S. AI models and fraudulently evaded protective measures under aliases and purchased dozens of accounts using a sophisticated network of international banking channels,” allowing them to mask identities and avoid detection.
OpenAI itself stated the following to the Select Committee regarding the matter:
“Through our review, we found that DeepSeek employees circumvented guardrails in OpenAI’s models to extract reasoning outputs, which can be used in a technique known as ‘distillation’ to accelerate the development of advanced model reasoning capabilities at a lower cost. Observations of DeepSeek’s R1 model also indicate instances of reasoning structures and phrase patterns that align with the behavior of OpenAI’s models. Additionally, we found that DeepSeek employees used OpenAI models to grade model responses and filter and transform training data, which are key steps in the AI development process. DeepSeek likely also used leading open-source AI models to create high-quality synthetic data.”
A Pattern of Scrutiny and Policy Recommendations
The committee’s report represents a new peak in U.S. government attention on DeepSeek. Previous actions included a bipartisan bill introduced Feb 6, 2025, aiming to ban DeepSeek from federal devices, along with earlier bans by the U.S. Navy and the state of Texas, plus international probes. Amidst this, DeepSeek has continued development, releasing efficient open-weight models and new AI alignment techniques.
Its efficiency focus attracted users like Tencent, which confirmed in March it was leveraging DeepSeek to reduce GPU needs, with an executive noting, “Chinese companies are generally prioritizing efficiency and utilization… DeepSeek’s success really sort of symbolize and solidify — demonstrated that — that reality.”
Based on its findings, the Select Committee proposed several policy recommendations. These include swift action to expand U.S. export controls to cover additional chips like the Nvidia H20 and enhancing enforcement through increased funding for the Bureau of Industry and Security (BIS), creating whistleblower incentives, requiring chipmakers to track end-users, installing on-chip location verification for restricted chip exports, and imposing remote access controls on data centers using U.S.-origin GPUs.
The committee also recommends further restricting the PRC’s AI development capability by expanding controls on semiconductor manufacturing equipment, placing a federal procurement ban on PRC-origin AI models for government devices, and directing agencies like NIST and CISA to develop security standards for AI developers to protect against risks like model distillation.
Finally, they call for better interagency coordination to monitor adversarial AI progress and prepare for potential strategic surprises in AI development.
Chairman Moolenaar encapsulated the committee’s view: “This report makes it clear: DeepSeek isn’t just another AI app — it’s a weapon in the Chinese Communist Party’s arsenal, designed to spy on Americans, steal our technology, and subvert U.S. law.” He added, “American innovation should never be the engine of our adversaries’ ambitions.”
