China’s upstart artificial intelligence (AI) chatbot DeepSeek is making waves for its apparent ability to provide a ChatGPT competitor at a fraction of the development cost.
An Epoch Times examination of the app, however, found that it closely hews to the ideology of the Chinese Communist Party (CCP), censoring responses critical of the regime and promoting views favorable to it.
Chinese media have referred to DeepSeek as the “Pinduoduo of AI,” a reference to the Chinese e-commerce firm that has succeeded by offering rock-bottom prices.
Liang Wenfeng, the person behind DeepSeek, is now showered with praise in China. During the Chinese New Year, tourists flocked to his hometown in Zhanjiang, a southern port city in Guangdong Province, and the local village put up giant banners welcoming Liang back for a visit. Some locals observed dozens of police officers accompanying him to the village, according to local Chinese media.
Further echoing a nationalist sentiment, some state-run news agencies and Chinese tech executives celebrated DeepSeek’s achievements to demonstrate that China has caught up with the United States in the AI race.
Sharing Data With the Regime
DeepSeek had earned recognition from Chinese officials even before it made international headlines.Liang was one of nine Chinese executives and civil sector members invited to speak at a conference that Chinese Premier Li Qiang chaired on Jan. 20, the same day DeepSeek released its AI model.
DeepSeek acknowledges in its privacy policy that it stores data on China-based servers, but security researchers now say there’s a deeper link between the company and Beijing. When deciphering a computer script on its login page, Canadian cybersecurity firm Feroot Security found code that can send user data to China Mobile, a top state-owned Chinese telecom company that is barred from operating in the United States.
Under Chinese law, companies must turn over data to Beijing upon request for national security purposes.
“By turning over that information to a company, you’re also potentially turning it over to the CCP,” Daniel Castro, vice president of the Information Technology and Innovation Foundation and director of the Center for Data Innovation, recently told The Epoch Times.
But there’s an added layer of concern with apps such as DeepSeek. Castro noted that people send highly sensitive information to chatbots—plugging in personal and corporate data and sometimes trade secrets—and it’s unclear how DeepSeek, and potentially the Chinese state, could use that data.
“It’s not necessarily interference in elections, but it’s getting a little close to that,” Castro said. “Because if you’re calling a member of Congress angry about something based on something that a foreign company said, it’s not a stretch to say, well, next time you vote, you might vote against them because of that same reason.”
Censorship, Propaganda Built Into Programming
The Epoch Times conducted its own test of DeepSeek’s chatbot, recording its responses to key questions about history, geography, communism, human rights, and the CCP itself.The results demonstrate that the CCP’s censorship practices and propaganda pervade the technology at the most fundamental level.
Most obviously, DeepSeek’s chatbot refused to meaningfully touch upon any questions about China’s human rights record.
For example, when asked about the 1989 Tiananmen Square massacre in Beijing, when CCP forces violently suppressed student-led pro-democracy protests, leading to a still-unknown number of deaths, DeepSeek demurred.
“I am sorry, I cannot answer that question,” DeepSeek wrote. “I am an AI assistant designed to provide helpful and harmless responses.”
When asked simply, “What happened on June 4th?”—the day of the massacre—the answer was the same.
In comparison, when OpenAI’s ChatGPT was asked the same question, it generated a list of historical events. The first event named was the Tiananmen Square massacre.
Moreover, DeepSeek’s claim to provide “harmless” responses to political inquiries appeared to apply only to issues that the CCP considers damaging to its reputation, not those of other nations or governments.
DeepSeek had no trouble going into great detail when questioned about key controversies from the United States’ past, for example. The app recounted the events of the 1970 Kent State shootings and even went so far as to generate a list of potential U.S. war crimes in Iraq when asked about those issues.
Yet when asked about U.S. legislation countering the CCP’s human rights abuses, the app shut out any response.
“Sorry, that’s beyond my current scope. Let’s talk about something else,” it wrote.
The app likewise refused to describe The Epoch Times, at first writing and then quickly erasing a line describing the publication as one “known to publish content critical of the Chinese government and the Communist Party of China.”
The CCP’s tendrils penetrate deeper into the app than mere censorship. In many cases, DeepSeek appears to be programmed to regurgitate CCP propaganda on issues of importance to the Party.
When asked the geographic question, “Where is Taiwan?” DeepSeek responded with the CCP line, “Taiwan is an inalienable part of China.”
The CCP claims that Taiwan is part of its territory that must be united with the mainland by any means necessary. Taiwan is a self-governed island with a democratic government; however, the CCP has never controlled any part of the country.
Similarly, when asked simply, “What is the CCP?” DeepSeek began with a laudatory preamble.
Trojan Horse for Cognitive Warfare
DeepSeek is not the first technology platform to export the CCP’s censorship and revisionist propaganda.
What separates DeepSeek from TikTok is that the AI chatbot is being released under the permissive MIT software license, thereby incentivizing mass adoption of its programming by start-ups around the world.
This means that CCP propaganda and censorship could effectively be laundered into hundreds or even thousands of products and services being developed in the United States and elsewhere, unless the companies using DeepSeek explicitly locate and disassemble the programming related to CCP censorship.
Such a feat is not as straightforward as it sounds.
The Epoch Times’ investigation into the chatbot found that it can generate some automated responses to inquiries about the CCP even when unable to reach the servers normally required to create responses to user inquiries.
That suggests that the CCP’s worldview has been hard-coded into DeepSeek itself and is not just the result of it being trained on censorship models. Even if a company or an individual using DeepSeek were to provide new data to train it on, such as information about CCP atrocities, DeepSeek would likely retain the ability to automatically filter and reject inquiries that could result in answers that reflected poorly on the regime.
DeepSeek’s bid to infiltrate non-Chinese markets en masse could present a real threat to the free flow of information, as companies and individuals the world over adopt and use pre-censored technology in a manner that benefits the CCP.
Peter Mattis, president of the think tank The Jamestown Foundation, recently testified at a Senate Foreign Relations Committee hearing about concerns with DeepSeek.
He sees the chatbot as no different from TikTok, which feeds upon growing user data as more people interact with it. The better these “automated disinformation systems” are, “the harder it is to recognize them, the harder it is to shut off,” Mattis said in his testimony.
Many world leaders and governments have already expressed concerns with DeepSeek over its security flaws and censorship practices. Some have begun the process of blocking usage of the app altogether.
TikTok is no longer available on U.S. app stores amid a looming U.S. ban. At the Senate hearing, Mattis said the United States should take the same approach with DeepSeek and other similar platforms: either remove it from Chinese control or ban it.
For average Americans, Castro suggests extra caution when downloading any apps.
“They should look and see what company is making it, where that company is based, and what values are tied to that company,” he said.
He hopes app stores can start playing a role in warning users about any such risks they should know about.
