Why the US Wants DeepSeek Off Government Systems & Why Deleting the App Is Only the Start

An official uploads a document. The chatbot returns a neat summary. A tedious job disappears in seconds. Where did the document go? That question sits at the centre of Washington’s move against DeepSeek. The concern extends beyond what the Chinese company’s AI can answer to what its services might collect, where information travels and who could gain access.

The most consequential part of the crackdown concerns technology embedded in other companies’ products. A government employee could encounter DeepSeek without opening an app bearing its name.

WHAT HAS THE US ACTUALLY ORDERED?

The Department of Homeland Security’s published order excludes DeepSeek and affiliated products and services from federal civilian procurement. It also requires removal of their products, alongside third-party products and services developed using DeepSeek implementations, from covered agency systems.

Its definition extends to model weights, applications and other implementations. Limited exceptions are available, including for specified research, testing and national-security purposes. This is a government systems and procurement restriction; it does not impose a general ban on Americans using DeepSeek privately.

WHY DOES WASHINGTON CONSIDER DEEPSEEK A SECURITY RISK?

DHS identifies several potential threats, including sensitive information reaching China, intelligence being assembled from government usage, and vulnerabilities or malicious functionality in software. Its reasoning combines technical concerns with the legal environment governing a China-based provider.

The underlying concern is straightforward: government employees routinely handle material whose importance may be invisible to an outsider.

A draft procurement document might reveal an impending purchase. A staffing spreadsheet might identify sensitive personnel. A series of seemingly ordinary prompts could expose an agency’s priorities.

These are illustrations of potential exposure, rather than documented incidents involving the order. They explain why a service acceptable for a casual query might face a much higher threshold before processing government information.

WHAT DOES DEEPSEEK ITSELF SAY ABOUT DATA?

DeepSeek’s privacy policy, updated February 10, 2026, says it may collect prompts, uploaded files, voice inputs, photographs, chat histories and account information. It also describes collecting device identifiers, IP addresses and usage information.

The policy says personal data is stored in China and may be used to develop and improve its technology, including training models. It describes circumstances involving disclosure to authorities and warns users against submitting sensitive personal information.

Those disclosures establish the company’s stated data practices. They do not establish that Chinese authorities have accessed a particular American official’s conversation.

That distinction separates a demonstrable exposure pathway from a proven act of espionage.

DID RESEARCHERS FIND ACTUAL SECURITY WEAKNESSES?

Yes, in an earlier version of DeepSeek’s iPhone application.

In research published on February 6, 2025, mobile-security company NowSecure reported unencrypted transmission of some registration and device information, weak encryption practices, insecure storage of credentials and extensive device-data collection. It also identified transmission to ByteDance-controlled infrastructure.

Some weaknesses could allow information to be intercepted or modified by an attacker with the necessary network access. Device information, accumulated over time, could also help identify users.

IS DEEPSEEK’S APP THE SAME THING AS ITS AI MODEL?

This is the distinction that makes the story bigger than an app ban.

The app is an interface through which someone uses a service. A model is the underlying system generating responses. DeepSeek has released downloadable models that other organisations can operate on their own infrastructure.

NowSecure itself distinguished the iPhone application from self-hosted deployments and versions hosted by other providers.

Technically, a properly isolated local deployment can process prompts without sending them to DeepSeek’s servers. That changes the data-exposure question. It does not automatically settle questions about the surrounding software, model behaviour, updates or access controls.

Crucially, the US order’s coverage of model weights means that keeping processing local does not, by itself, place a covered deployment beyond the restriction.

WHY DOES THE THIRD-PARTY CLAUSE MATTER SO MUCH?

Because the name on a product can conceal the technology underneath it.

Imagine a supplier selling a document assistant under its own brand. The interface, contract and customer support could all belong to that supplier, while the underlying model comes from DeepSeek.

For a buyer, checking the product’s logo would reveal very little about its dependencies.

The practical implication is a deeper procurement question: which models power this service, who operates them, and can the supplier change them after the contract is signed?

That is an analytical consequence of the order’s breadth. Government buyers will need visibility into their AI suppliers’ suppliers.

HAS THE US PROVED THAT DEEPSEEK IS A SPYING OPERATION?

The published material supports a security-risk assessment and a decision to restrict government exposure. That is different from publicly demonstrating a completed espionage operation.

Separately, an April 2025 investigation by the House Select Committee on the Chinese Communist Party accused DeepSeek of transferring American user data to China and producing responses shaped by Chinese censorship requirements. Those are findings and allegations attributable to the committee.

Data security and answer manipulation also require separate scrutiny. One concerns who can obtain information; the other concerns what the system tells its user.

HOW HAS CHINA RESPONDED?

Beijing has rejected the American action.

According to Reuters on October 20, China’s commerce ministry called for Washington to reverse discriminatory restrictions, argued that the move disregarded understandings between the two countries’ leaders and warned that China would protect its interests.

The disagreement therefore combines a security dispute with a commercial contest. Government purchasing restrictions can affect a technology provider’s reputation and prospects beyond the contracts immediately covered.

WHAT DOES THIS MEAN FOR INDIA?

India has already confronted the confidentiality question.

Reuters reported in February 2025 that the finance ministry asked employees to avoid AI tools including ChatGPT and DeepSeek for official purposes because of risks to government documents and data. That advisory covered tools from both American and Chinese companies.

For Indian institutions, the useful questions begin with the actual deployment: where prompts are processed, whether they are retained, whether they can be used for training, which subcontractors receive them and who controls access.

Hosting a model in India can change some risks. Establishing trust still requires examining the complete system. The DeepSeek dispute exposes a weakness in how AI is often purchased: buyers test the quality of the answer before investigating the journey of the question.

With inputs from ANI & agencies