How DeepSeek Censorship Works and How to Get Around It
A WIRED investigation shows that the popular Chinese AI model is censored on both the application and training level. While the firm seems to have an edge on US rivals in terms of math and reasoning, it also aggressively censors its own replies. Ask DeepSeek R1 about Taiwan or Tiananmen, and the model is unlikely to give an answer.
Application-Level Censorship and Regulations
After DeepSeek exploded in popularity in the US, users who accessed R1 through DeepSeek’s website, app, or API quickly noticed the model refusing to generate answers for topics deemed sensitive by the Chinese government. These refusals are triggered on an application level, so they’re only seen if a user interacts with R1 through a DeepSeek-controlled channel. Rejections like this are common on Chinese-made LLMs because a 2023 regulation on generative AI specified that AI models in China are required to follow stringent information controls that also apply to social media and search engines. The law forbids AI models from generating content that “damages the unity of the country and social harmony.”
“DeepSeek initially complies with Chinese regulations, ensuring legal adherence while aligning the model with the needs and cultural context of local users,” says Adina Yakefu, a researcher focusing on Chinese AI models at Hugging Face. Because R1 is a reasoning model that shows its train of thought, this real-time monitoring mechanism can result in the surreal experience of watching the model censor itself as it interacts with users. When WIRED asked R1 about sensitive topics, the model first started compiling a long answer; yet shortly before it finished, the whole answer disappeared and was replaced by a terse message: “Sorry, I'm not sure how to approach this type of question yet. Let's chat about math, coding, and logic problems instead!”
Technical Methods to Bypass Filters
The fact that R1 is open source means there are ways to get around the censorship matrix. WIRED found that while the most straightforward censorship can be easily avoided by not using DeepSeek’s app, there are other types of bias baked into the model during the training process. To figure out how this works on a technical level, WIRED tested DeepSeek-R1 on its own app, a version of the app hosted on a third-party platform called Together AI, and another version hosted on a WIRED computer, using the application Ollama.
- Local Execution: You can download the model and run it locally, which means the data and the response generation happen on your own computer.
- Distilled Versions: DeepSeek has smaller, distilled versions that can be run on a regular laptop.
- Cloud Infrastructure: You can rent cloud servers outside of China from companies like Amazon and Microsoft. This work-around is more expensive and requires more technical know-how than accessing the model through DeepSeek’s app or website.
DeepSeek-R1 Access and Censorship Comparison
| Method | Censorship Context | Technical Setting |
|---|---|---|
| DeepSeek App / Website | Refuses to answer topics deemed sensitive | DeepSeek-controlled channel |
| Together AI | Straightforward censorship can be easily avoided | Third-party platform |
| Ollama (Local) | Bypasses application-level refusals | Run on your own computer |
| Amazon / Microsoft | Filters can be avoided | Cloud servers outside of China |
These findings have major implications for DeepSeek and Chinese AI companies generally. If the censorship filters on large language models can be easily removed, it will likely make open-source LLMs from China even more popular, as researchers can modify the models to their liking. If the filters are hard to get around, however, the models will inevitably prove less useful and could become less competitive on the global market.