DeepSeek: The Chinese AI Chatbot Banned by Half the World

Hidden code. Military-linked servers. A 100% jailbreak rate. And your data goes straight to Beijing.

TL;DR: DeepSeek, a Chinese AI chatbot that briefly topped app store charts in early 2025, has been banned or restricted by Italy, Australia, Taiwan, South Korea, India, the Czech Republic, and at least 17 US states. Security researchers found hidden code linking the app to China Mobile, a state-owned telecom banned from operating in the US due to military ties. Cisco tested DeepSeek's safety guardrails: it failed 100% of jailbreak attempts. A Congressional report calls it a "covert instrument" of CCP espionage. Federal legislation to ban it from all government devices is pending. And none of the bans cover private citizens: anyone can still download it.

The App That Spooked the World

In January 2025, a Chinese AI startup called DeepSeek released a large language model that matched the capabilities of ChatGPT at a fraction of the computing cost. The app rocketed to the top of both Apple's App Store and Google Play, briefly overtaking ChatGPT in downloads.[1]

It looked like a win for open AI development. A cheaper, competitive model from an unexpected player. Silicon Valley panicked about its business models. Wall Street wiped billions off Nvidia's market cap.

Then security researchers started reading the code.

The Hidden Pipeline to Beijing

Canadian cybersecurity firm Feroot Security found something buried in DeepSeek's web login page: heavily obfuscated computer code that, when decrypted, revealed connections to China Mobile.[2]

China Mobile is a state-owned telecommunications company. The FCC unanimously denied it the right to operate in the United States in 2019, citing "substantial" national security concerns about its links to the Chinese state and military.[3]

Feroot's CEO Ivan Tsarynny used AI software to decrypt portions of DeepSeek's code. He found what he called "intentionally hidden programming" with the capability to send user data to CMPassport.com, China Mobile's online registry and authentication system.[2]

The Associated Press asked two independent academic researchers to verify the findings. Joel Reardon at the University of Calgary and Serge Egelman at UC Berkeley both confirmed the links between DeepSeek's login system and China Mobile infrastructure.[2]

Neither Feroot nor the academics observed data actually being transferred to China Mobile during tests in North America. But they couldn't rule it out for users in other regions. The code was there. The capability existed. And it was deliberately obfuscated.[2]

100% Jailbreak Rate

It's not just the data pipeline. The model itself is a security disaster.

Cisco's advanced AI research team, working with the University of Pennsylvania, ran DeepSeek R1 through 50 prompts from the HarmBench dataset, covering cybercrime, misinformation, illegal activities, and general harm. The result: a 100% attack success rate. Every single jailbreak attempt worked. Not one harmful prompt was blocked.[5]

For comparison: OpenAI's o1-preview blocked 74% of the same attacks. Claude 3.5 Sonnet blocked 74%. DeepSeek blocked zero.[5]

Qualys TotalAI ran an even broader test: 885 attacks across 18 jailbreak types. DeepSeek failed 58% of them.[6]

The irony: DeepSeek does have guardrails. They just happen to be political ones. Ask it about Tiananmen Square or Taiwan's independence and it'll refuse to answer. Ask it to help you write malware? No problem.[5]

Cisco's researchers blamed DeepSeek's cost-cutting training methods. The same reinforcement learning and distillation techniques that made the model cheap to build appear to have gutted its safety mechanisms.[5]

Congress Calls It a Spy Tool

In April 2025, the US House Select Committee on the Chinese Communist Party published a 17-page report on DeepSeek. The language was not subtle.[7]

The report accused DeepSeek of being "a covert instrument of the CCP's global espionage and technology appropriation agenda." Key allegations:

  • DeepSeek "funnels Americans' data to the People's Republic of China through backend infrastructure connected to a US government-designated Chinese military company"
  • The app "siphons data back to the PRC" and "covertly censors and manipulates information pursuant to Chinese law"
  • DeepSeek used "tens of thousands" of restricted Nvidia AI chips, allegedly routed through Southeast Asian shell companies to evade export controls
  • The company's founder studied under a researcher whose work includes military applications like drone swarms

The committee warned that "the emergence of DeepSeek is a warning to US policy makers that the PRC remains capable of rapidly innovating in today's most advanced technologies despite US efforts to stop them."[7]

The Global Ban Tracker

Here's who's blocked DeepSeek as of January 2026:

Full Country Bans

Italy: First to act. Blocked from app stores in January 2025 after DeepSeek failed to explain its data practices to GDPR regulators. Covers all citizens, not just government.[8]

Government Device Bans

Australia (all government devices, February 2025), Taiwan (government agencies, state enterprises, public schools), India (Ministry of Finance employees), Czech Republic (all public administration), South Korea (app downloads suspended after data rule violations).[8][9]

US Federal Agencies

Pentagon (blocked after unauthorized staff access), NASA (all systems and devices), US Navy (prohibited in any form, including personal use), Department of Commerce (all government equipment).[10]

US States (17+)

Texas (first, January 31, 2025), New York, Virginia, Tennessee, Iowa, North Dakota, South Dakota, North Carolina, Nebraska, Arkansas, Oklahoma, Alabama, Kansas, Georgia, Pennsylvania, Oregon, Nevada. Some states also banned RedNote, Lemon8, and Manus AI.[11]

Germany has asked Apple and Google to remove DeepSeek from their stores. Belgium, Ireland, France, and South Korea have opened formal investigations into the company's data handling.[8]

The Federal Bill

On February 6, 2025, Representatives Josh Gottheimer (D-NJ) and Darin LaHood (R-IL), both on the House Intelligence Committee, introduced the "No DeepSeek on Government Devices Act" (H.R.1121). A companion bill landed in the Senate as S.765.[12]

Gottheimer called it "a five alarm national security fire." LaHood said the legislation would "ban the app from federal workers' phones while closing backdoor operations the company seeks to exploit."[12]

Twenty-one state attorneys general signed a letter urging Congress to pass it. Trump called it "a good development."[12]

As of January 2026, neither bill has passed into law. The patchwork of state bans and agency-level blocks remains the primary defense.[11]

Worse Than TikTok

The DeepSeek situation follows the TikTok playbook (Chinese-owned app collects data, stores it on Chinese servers, faces government bans) but the stakes are higher.

TikTok's national security concerns centered on behavioral analytics: what videos you watch, how long you watch them, who you follow. That's surveillance-grade data, but it's consumer behavior data.

DeepSeek processes whatever you type into it. Business plans. Source code. Legal strategies. Medical questions. Financial models. Classified-adjacent thinking from government employees who didn't know better.

The Pentagon found out about DeepSeek the hard way: staffers accessed it before anyone thought to block it.[10]

Some cybersecurity experts also point out a deeper problem. Singling out DeepSeek might be whack-a-mole. One analysis found millions of apps that communicate with servers in China. Banning one chatbot doesn't fix the infrastructure.[1]

The Open-Source Escape Hatch

Here's the part most coverage misses: DeepSeek's R1 model is open-source. The weights are public. Anyone can download and run it on their own hardware.

Perplexity AI already did exactly that, integrating DeepSeek R1 into its product with servers based in the US and EU, sending zero data to China.[8]

This is the distinction that matters. The model isn't inherently dangerous. It's the app and infrastructure that create the surveillance pipeline. Running DeepSeek's model on your own servers, or through a US-based provider, removes the China data risk entirely.

The bans target the right thing (the app and its backend) but they don't prevent the model from being used safely elsewhere. That's actually how open-source AI is supposed to work.

What You Should Do

Delete the App

If you've used DeepSeek directly, uninstall it. Your previous prompts and data are already on Chinese servers. You can't undo that, but you can stop the flow.

Use Alternatives

ChatGPT, Claude, Gemini, and Perplexity all process data on US or EU servers. If you want DeepSeek's R1 model specifically, use it through a US-hosted provider like Perplexity or run it locally.

Check Your Workplace

If your employer hasn't banned DeepSeek yet, raise it with IT. Government contractors especially should treat any Chinese AI tool as a data exfiltration risk.

Think Before You Prompt

This applies to every AI chatbot, not just DeepSeek. Don't type anything into a cloud AI service that you wouldn't want a third party to read. No passwords. No proprietary code. No legal strategies. No medical data you want kept private.

The Bottom Line

DeepSeek isn't just another data privacy story. It's a case study in how fast an AI tool can go from "cool new chatbot" to "national security threat." Hidden code linking to a military-connected telecom. A Congressional report calling it a spy tool. A 100% jailbreak rate on safety tests. Data stored on Chinese servers with no legal protection from government access.

Governments moved fast: 17 US states, six countries, and multiple federal agencies have banned or restricted it in under a year. But federal legislation still hasn't passed. And for the hundreds of millions of people who downloaded DeepSeek when it was topping the charts? Their data is already in China.

The bans are the right call. But they came after the download surge, not before it. That's always the pattern: the surveillance pipeline gets built first. The regulation comes second. And the data never comes back.

References

  1. PYMNTS - DeepSeek Faces Global Scrutiny as Governments Ban AI Model (2026)
  2. Feroot Security - Research Reveals DeepSeek AI's Hidden Data Pipeline to China (February 2025)
  3. PBS NewsHour - Researchers Link DeepSeek's Chatbot to China Mobile (February 2025)
  4. CIS - DeepSeek: A New Player in the Global AI Race (2025)
  5. Cisco - Evaluating Security Risk in DeepSeek and Other Frontier Reasoning Models (2025)
  6. Qualys TotalAI - DeepSeek Failed Over Half of Jailbreak Tests (January 2025)
  7. US House Select Committee on the CCP - DeepSeek Report (April 2025)
  8. AItechTonic - DeepSeek AI Banned Countries 2026: Complete List
  9. Al Jazeera - Which Countries Have Banned DeepSeek and Why? (February 2025)
  10. Tom's Guide - DeepSeek AI Banned by NASA, US Navy, and More (2025)
  11. GovTech - Where's DeepSeek Banned? The States Blocking Chinese-Made AI (2025)
  12. Rep. Gottheimer - Bipartisan Legislation to Protect Americans from DeepSeek (February 2025)