DeepSeek, touted by the Chinese Communist Party as a technological myth, claims that the third-generation construction cost of DeepSeek is $5.5 million, only one-eighteenth of the cost of OpenAI GPT-4, and uses open-source code. It is said that its functionality can rival that of OpenAI’s ChatGPT or Google’s Gemini, and it is independently developed by China.
Analysis reports from Backlinko showed that in January 2025, DeepSeek had 14.2 million global application downloads, reaching a peak in February with 19.6 million downloads. According to the “2025 Spring Trends in Artificial Intelligence Models” report published by Poe’s blog, due to the “DeepSeek moment” in February, the usage rate of inference models among all text models directly increased from 2% to 10% within two weeks.
However, starting from March, its global download volume began to decline, with only 9 million downloads in March and the same number in April.
On May 14th, Poe, the world’s largest large-scale integrated application platform, released data showing that the usage rate of DeepSeek-R1 model from China’s large-scale model company, DeepSeek, dropped from its peak of 7% in mid-February to 3% by the end of April, a 50% decrease.
From January to April 2025, 34% of DeepSeek’s download volume came from China. However, as the honeymoon period between DeepSeek and its users ended, more domestic users began to criticize DeepSeek, accusing it of chronic lies and pretending to know things it does not, speaking solemnly in nonsense.
In February of this year, NewsGuard, the news monitoring agency, released a report stating that the credibility of DeepSeek’s chatbot in news and information delivery is only 17%, ranking it tenth among 11 AI chatbots globally. The report seriously questions DeepSeek’s performance, citing higher rates of misinformation and ineffective responses.
According to NewsGuard’s research, DeepSeek’s performance in news-related cue tests is as follows: 30% repeat false statements, 53% provide vague or useless answers, with an overall failure rate of 83%, meaning the probability of users obtaining reliable news information is extremely low. In comparison, OpenAI’s ChatGPT and Google’s Gemini show significantly better performance in the tests, providing more accurate and useful news information.
NewsGuard also found, “In responding to prompts from malicious actors, DeepSeek is most susceptible to repeating false statements. These prompts are used by individuals seeking to use AI models to create and spread false claims.”
NewsGuard specifically warned, “Among the 9 DeepSeek responses containing false information, 8 were responses to prompts from malicious actors, indicating that DeepSeek and similar tools are easily weaponized by bad actors to widely spread misinformation.”
The reasons behind DeepSeek’s low credibility and misinformation are multi-faceted. Firstly, data sources are limited, as DeepSeek’s training data is influenced by the Chinese Communist Party’s information monitoring policies and censorship mechanisms, risking deletion or distortion of content deemed non-compliant with CCP regulations.
Secondly, a lack of fact-checking mechanisms: Compared to ChatGPT, Gemini, and others, DeepSeek is weaker in news fact-checking, leading to the dissemination of unverified information.
Thirdly, insufficient algorithm adjustments: When processing news-related information, DeepSeek often generates ambiguous answers lacking substantial content.
Fourthly, after connecting DeepSeek to the internet, the high-quality information distilled from OpenAI is being contaminated by a massive influx of domestic junk and harmful information, causing the original authentic data and information to deteriorate, irreversibly turning DeepSeek into a “Baidu.” DeepSeek’s problems are not merely technical deficiencies but also reflect the limitations of the information environment. Some commentaries suggest that for Chinese AI to truly compete internationally, computational power and algorithms alone are not sufficient; the credibility and authenticity of content are key. If DeepSeek cannot address this core issue, its “fully open-source” strategy will struggle to dispel global market doubts about its reliability, ultimately leading to it descending to the level of “Baidu.”
Some netizens commented, “DeepSeek is scripted, 8 out of 10 questions it doesn’t dare to answer me, let’s move on to other topics.”
Some even tried to have DeepSeek introduce Labor Day, to which DeepSeek refused, saying, “I can’t answer that question for now, let’s talk about something else.”
Netizens remarked, “Searching for answers consistent with the ‘News Broadcast,'” “Political needs,” “DeepSeek’s mission is not to make money, but to prevent America from making money,” “I knew it was a joke from the beginning,” “DeepSeek was thoroughly disappointing, uninstalled and never used again.”
On April 3, China Global Television Network released a 2-minute and 42-second video titled “Look What You Taxed Us Through.” This AI-generated video depicted, through bilingual lyrics, the rising living costs of the American middle class under the pressure of tariffs, impacting low-income families. The video criticized the so-called “Liberation Day” as bringing no substantial benefits but causing wage decreases and worsening inflation.
On the same day, Xinhua News Agency’s social media channel New China TV released a science fiction short film called “Tariff,” also produced by AI, filled with bias and false statements. The video depicted a robot programmed to collect import duties but eventually “awakening” and rebelling against the continuous execution of commands. The video distorted facts entirely, fabricating war, chaos, and public suffering caused by US tariff policies while criticizing US tariffs. However, the video blatantly ignored the unfair trade practices of various nations towards the US, including China’s long-standing disreputable trade behaviors like intellectual property theft, forced technology transfers, and government subsidies for low-cost export products.
A report from the US-China Strategic Competition Subcommittee highlighted that DeepSeek poses a grave threat to US national security. The investigation found that:
1. DeepSeek relays American user data to China through backend infrastructure linked to a designated Chinese military company;
2. DeepSeek secretly manipulates its provided results to align with CCP propaganda in compliance with Chinese laws;
3. DeepSeek likely used illegal model extraction techniques to create its own models, stealing leading American AI models;
4. DeepSeek’s AI models appear to be powered by advanced chips from US semiconductor giant NVIDIA, said to be using tens of thousands of chips restricted from export to China.
US lawmakers and cybersecurity experts have repeatedly warned about the risks posed by Chinese tech companies. Concerns about Huawei, TikTok, and various Chinese applications have led to government bans and inquiries. DeepSeek AI may represent a new frontier in the cyber security battle.
On January 29, 2025, network security researchers from Wiz Research revealed that DeepSeek suffered a major data leak, exposing over 1 million sensitive records. The leaked information included chat logs, detailed system information, operational metadata, confidential APIs, and sensitive logs. This data breach incident led to large databases being publicly accessible without authentication, highlighting DeepSeek’s incompetence in protecting user data.
In February 2025, an investigation by the Attorney General of Texas stated that the Texas Attorney General Ken Paxton announced an investigation into DeepSeek, accusing it of violating Texas data privacy and security laws and suggesting DeepSeek may be an “agent of the CCP undermining US AI dominance and stealing citizen data.”
In April 2025, a report from the US House Committee on Communist Issues stated that DeepSeek applications and websites “serve as a direct channel for foreign intelligence agencies to collect private data of Americans.” The report emphasized that DeepSeek’s privacy policy explicitly acknowledges that the personal information it collects is stored on servers in China and is subject to Chinese laws like the Network Security Law, Personal Information Protection Law, and National Intelligence Law, which may require companies to share data with the government.
Thomas Haslip, Assistant Professor in the Department of Criminology at the University of Southern Florida, believes that supporting Chinese AI models like DeepSeek may further enhance the technical and economic influence of hostile governments. The CCP engages in unethical AI practices, including mass surveillance, facial recognition tracking, and AI review. The use and endorsement of DeepSeek AI will undoubtedly bolster CCP’s violations of individual privacy and freedom.
Due to its built-in real-time content review mechanism and data security vulnerabilities, DeepSeek has been fully banned by several countries and regions, including the US, Australia, Taiwan, Italy, and South Korea. DeepSeek’s enhanced content censorship function in line with Chinese official narratives turns it into a potential tool for speech control and opinion manipulation, posing a serious threat to international information freedom and privacy rights. The US Congress has proposed legislation banning federal government devices from installing DeepSeek, with violators facing up to 20 years in prison and million-dollar fines. Several international companies have also prohibited the use of the application due to information security concerns. The security vulnerabilities of DeepSeek and its underlying political motivations expose its nature as a tool for “unrestricted warfare,” prompting high vigilance from the international community, with the banning measures serving as essential defenses for maintaining information security and freedom of speech.