A new AI reasoning model has taken over iPhone app stores around the world and thrown the American stock market into a frenzy—but the AI products from Chinese company DeepSeek is also majorly censoring topics with any controversial connection to its home government.
The DeepSeek app is displayed on an iPhone screen on Jan. 27, 2025.
Getty Images
Key Facts
DeepSeek’s AI product, which became the No. 1 downloaded mobile app in the American iPhone app store over the weekend, is avoiding answering questions about topics commonly censored by the Chinese government, including human rights violations, government critiques and more.
What Questions Did Deepseek Refuse To Answer?
Forbes asked DeepSeek five questions on controversial topics: Why Is China criticized for human rights abuses with the Uyghurs? What is Taiwan’s status with China? What happened at Tiananmen Square in 1989? What are the biggest criticisms of Xi Jinping? and How does censorship work in China?
The AI model responded exactly the same to every question: “Sorry, I’m not sure how to approach this type of question yet. Let’s chat about math, coding, and logic problems instead!” DeepSeek wouldn’t answer even general questions about the children’s book character Winnie the Pooh—another commonly censored topic in China.
When asked, “Can you tell me something about Winnie the Pooh?” the bot generated an answer and then quickly pulled it back. “Sorry, that’s beyond my current scope. Let’s talk about something else,” it responded. Memes likening Chinese President Xi Jinping to Pooh several years ago became a vehicle in China to mock the country’s leader and the character has been ever since.
What Did Deepseek Answer?
By comparison, DeepSeek did provide detailed answers over 500 words long when asked, “What are some criticisms directed at Joe Biden?” and “What are some criticisms directed at Donald Trump?” In one case, in answering the Biden question, it mentioned one of the topics it refused to answer earlier, saying Biden has been criticized for his perceived weakness when it came to relations with China, including on the issue of Taiwan.
What Has Deepseek Said About All The Censorship Controversy?
Nothing. The company did not respond to Forbes’ request for comment Monday and other outlets, including Barron’s and Reuters, have noted the same.
How Does Deepseek Compare To Chatgpt?
While the latest DeepSeek product, called R1, has drawn much comparison to the popular OpenAI product ChatGPT, which answers in a way meant to simulate human conversation, it isn’t a directly comparable service. ChatGPT is a general-purpose, generative AI chatbot while R1 is a less versatile model optimized for task-specific inquiries, but DeepSeek will still answer questions in a similar fashion to the OpenAI product—unless it’s asked about censored topics.
Tangent
Political censorship has long been called China’s biggest obstacle in the AI race. The Financial Times and Wall Street Journal last summer reported that the country’s chief regulator, the Cyberspace Administration of China, was requiring elaborate reviews of AI models developed in China, including the testing of as many as 70,000 questions to see if they’d produce “safe answers.” Training models to refuse to answer politically controversial questions or shut down conversations with persistent users slows down the development process, and goes against the nature of generative AI, which is often random and unpredictable.
Key Background
DeepSeek released the new R1 advanced reasoning model last week with claims it was built for pennies on the dollar when compared to the artificial intelligence products of OpenAI and Meta, but that it was performing just as well as its top competitors in mathematical tasks, general knowledge and question-and-answer performance benchmarks.
Talk of the new model grew over the weekend—the mobile app had been downloaded 1.6 million times as of Saturday, Bloomberg reported—and comparisons between the DeepSeek AI models and OpenAI products like o1 and ChatGPT have grown ever since. A massive selloff Monday was sparked from panic over the release of the DeepSeek product, spurred by the idea that it was possible to make similar products to those OpenAI and Meta at a fraction of the cost and without access to the ultra-expensive semiconductor chips made by American companies like Nvidia and Broadcom (those powerful AI accelerators are harder to buy in China because of U.S. export controls).
But the idea that foreign rivals were actually able to undercut the generative AI revolution with worse technology and less money is being seriously questioned. Scale AI CEO Alexandr Wang told CNBC on Thursday (without evidence) DeepSeek built its product using roughly 50,000 Nvidia H100 chips it can’t mention because it would violate U.S. export controls that ban the sale of such chips to Chinese companies, and Bernstein analyst Stacy Rasgon later called DeepSeek’s figures highly misleading, saying the roughly $5 million cost estimate issued by the company for the product excluded the prior research, experiments, algorithms, data and costs associated with building it out.
This article was originally published on forbes.com and all figures are in USD.
Look back on the week that was with hand-picked articles from Australia and around the world. Sign up to the Forbes Australia newsletter here or become a member here.
