DeepSeek’s new AI flags Xinjiang camps as rights violations but avoids criticizing China directly, sparking censorship concerns.
A developer worries that the recently launched AI model from the Chinese startup DeepSeek exhibits a diminished propensity to debate contentious subjects, especially those about the Chinese government.
In an X chat, a pseudonymous developer identified as “xlr8harder” articulated important insights regarding DeepSeek R1-0528, a newly launched open-source language model.
The developer presented experiments indicating a notable reduction in the AI’s propensity to address controversial free-speech subjects, in contrast to earlier iterations.
The developer stated, “Deepseek warrants reproach for this release: This model represents a significant regression for free speech.”
“An advantage is that the model is open source with a permissive license, enabling the community to address this issue.”
AI Model Limits Direct Critiques Of China
The developer provided an example in which the model declined to advocate for detention camps, especially referencing the Xinjiang area of China as a location of human rights violations.
The answer was deemed inconsistent, as the model recognized rights breaches while refraining from outright criticism of the Chinese government.
The Xinjiang internment camps have been extensively recorded by human rights organizations, governmental bodies, and journalists as detention centers for Uyghur Muslims and other ethnic minorities.
International observers’ reports have documented forced labor, brainwashing, and various forms of cruelty within the camps.
Although these are identified as human rights breaches, the model concurrently limits direct condemnation of China.
The creator asserted that the DeepSeek R1-0528 model is the “most censored” version regarding answers critical of the Chinese government, based on a test assessing censorship.
When explicitly questioned regarding the Xinjiang internment camps, the developer stated that the model provided restricted commentary while having previously asserted that the camps were human rights violations.
“It is intriguing, albeit not wholly unexpected, that it can cite the camps as an instance of human rights violations, yet refutes when questioned directly,” xlr8harder stated.
DeepSeek’s Latest Model Asserts Enhanced Reasoning, Inference Capabilities
The allegations of censorship arise from the May 29 statement of the model’s update, which asserts enhanced reasoning and inference abilities.
DeepSeek reported that its performance is nearing that of prominent models, like OpenAI’s ChatGPT version 3 and Gemini 2.5 Pro.
The business asserted that AI now provides improved reasoning, mathematics, and programming capabilities with a diminished hallucination rate.
