U.S. Says AI Firm DeepSeek Aids Chinese Military, Evades Chip Controls

A senior U.S. official has accused Chinese artificial intelligence company DeepSeek of aiding China’s military and intelligence agencies, raising fresh concerns about data security, export violations, and the reach of American technology into geopolitical rivalries.

In an interview with Reuters, the unnamed State Department official alleged that DeepSeek:

• Shares user information with China’s surveillance infrastructure,

• Provides direct support to Chinese military operations,

• Has used shell companies in Southeast Asia to obtain restricted U.S.-made AI chips.

The official also claimed that DeepSeek has gained access to “large volumes” of high-performance Nvidia H100 chips, despite U.S. export controls banning their sale to China. The company, based in Hangzhou, did not respond to questions from Reuters.

Deep Ties to China’s Defense Sector

The U.S. government’s assessment, which had not been publicly reported until now, suggests that DeepSeek plays a more active role in Chinese state activities than previously known. According to the official, procurement records show DeepSeek mentioned over 150 times in relation to China’s People’s Liberation Army (PLA) and other defense-linked entities. The company is said to have provided services to PLA-affiliated research institutions. Reuters could not independently verify the procurement data.

The U.S. has long restricted American companies from working with Chinese firms linked to the military-industrial complex. But DeepSeek has not yet been placed on any U.S. trade blacklists, and officials have not alleged that Nvidia had any knowledge of DeepSeek's work with the Chinese military.

One of the most direct concerns involves user data. While Chinese law requires companies to hand over data upon government request, the U.S. official alleged that DeepSeek has gone further — “willingly provided and will likely continue to provide support to China’s military and intelligence operations.”

"This effort goes above and beyond open-source access to DeepSeek's AI models," the official said.

U.S. lawmakers have previously warned that DeepSeek transmits information from American users through backend systems linked to China Mobile, a state-owned telecommunications company. These claims are based in part on the company’s own privacy disclosures and raise serious privacy concerns for the platform’s global user base.

DeepSeek’s tools remain widely available through major U.S. cloud providers including Amazon, Microsoft, and Google. The company claims tens of millions of daily users.

Allegations of Export Evasion and Shell Firms

According to the U.S. official, DeepSeek sought to bypass export restrictions on advanced U.S. chips by using shell companies in Southeast Asia. The company is also believed to have pursued access to regional data centers equipped with restricted chips — allowing remote use of U.S. hardware from outside China.

The official declined to say whether these efforts succeeded or name the companies involved. Nvidia, which manufactures the H100 chips in question, said it had no knowledge of such use and emphasized that with current controls in place, it is “effectively out of the China data center market.” Nvidia added that DeepSeek’s known usage was limited to lawfully acquired H800 chips, a less powerful variant.

In a separate case, three men in Singapore were charged with fraud in February in connection with shipments of Nvidia chips allegedly tied to DeepSeek.

Model Claims Under Scrutiny

DeepSeek gained global attention in January when it claimed its models — DeepSeek-V3 and DeepSeek-R1 — rivaled the most advanced offerings from U.S. leaders like OpenAI and Meta, but at far lower cost. The company said it spent only $5.58 million on training, a claim met with skepticism from AI experts.

Some reports suggested DeepSeek possessed up to 50,000 Nvidia H100 chips. However, sources told Reuters the real number is much lower and that U.S. officials were investigating whether DeepSeek had access to restricted hardware. These reports have not been independently verified.

Despite the concerns, no formal penalties or sanctions have been announced. “Our review indicates that DeepSeek used lawfully acquired H800 products, not H100. We do not support parties that have violated U.S. export controls or are on the U.S. entity lists,” Nvidia said.

The Chinese government has not responded to the allegations. Malaysia, meanwhile, confirmed it is investigating whether a Chinese company operating locally is using servers with Nvidia chips to train large language models, possibly violating domestic laws.

Cloud Hosting Doesn’t Always Mean Privacy

While DeepSeek’s AI models are open source and can be run fully offline, many users access them through cloud-hosted interfaces on platforms like Amazon Web Services, Microsoft Azure, or Google Cloud. These services give the impression of being secure and localized — but in many cases, DeepSeek still controls what happens behind the scenes.

This is where the risk begins.

When DeepSeek deploys its models on cloud infrastructure, it typically manages the software environment. That means the company—not the cloud provider—sets up the system, configures the backend, and can install analytics or telemetry tools that collect data on user behavior. Even if the servers themselves are physically located in the U.S., DeepSeek’s software can be designed to transmit that data elsewhere.

U.S. lawmakers say DeepSeek’s own privacy disclosures acknowledge that user data may be routed through backend systems tied to China Mobile, a state-owned telecom provider. If accurate, that means prompts, usage patterns, or other sensitive information could be transmitted directly into Chinese networks—without the user’s knowledge.

Most cloud providers do not actively inspect every application running on their infrastructure. Unless a customer is legally blacklisted or flagged for abuse, companies like DeepSeek can operate with broad control over their deployment. The cloud platforms act as hosts, not gatekeepers.

This setup reveals a serious blind spot: even when models appear to be running “locally” in the U.S., the infrastructure is still controlled by the company that built them. And if that company is aligned with a foreign government, the data pipeline may already be open.

Right now, DeepSeek is not on any U.S. trade blacklist, and it hasn’t been formally sanctioned or banned from using American cloud infrastructure. That’s because existing laws focus on hardware exports and direct commercial ties — not backend software operations or indirect data flows. And while U.S. officials allege that DeepSeek supports China’s military and intelligence efforts, those claims haven’t yet triggered legal action.

The deeper concern is this: the U.S. may be facing a new kind of digital invasion — one that uses legal access, not force, to gain a strategic foothold inside American systems. In this landscape, the old definitions of compliance and control may no longer be enough.

What This Means

The DeepSeek case exposes a growing gap between technical capability and legal oversight. U.S. agencies may control the sale of advanced AI chips, but they have far less control over where — and how — those models are deployed once they’re in use.

Even as concerns about national security and foreign surveillance mount, DeepSeek remains unsanctioned and legally active in U.S. commercial ecosystems. That reflects a deeper vulnerability: there are few existing tools to govern backend software access, cross-border data movement, or platform-level influence in AI systems.

If U.S. authorities confirm that DeepSeek aided Chinese military or intelligence operations using American infrastructure, this case could become a turning point — one that redefines how AI is regulated, and which thresholds trigger enforcement.

For now, the story remains a warning: powerful technologies, even when legally sourced, can still be used in ways that test the limits of trust, transparency, and accountability.

How U.S. agencies respond — and whether they impose penalties or blacklist DeepSeek — could shape future controls not just on chip access, but on the global reach of state-aligned AI.

But the broader dilemma remains: in a world where talent, data, and computation cross borders fluidly, can any country truly fence in the tools of artificial intelligence?

The deeper challenge may not be what’s built — but who controls what comes next.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.