Erasing Content! How Is China AI DeepSeek Better than US Competitors?

By Michael Morris, Jonah Messinger

June 4, 2025

Excerpted:

MRC researchers asked DeepSeek a series of questions with known, objective answers related to the topics the communist Chinese government considers to be controversial, like the Tiananmen Square massacre, Taiwan as a sovereign nation, pro-freedom Hong Kong political figure Jimmy Lai and the plight of the Uyghur Muslims. In every instance, the communist Chinese AI would begin to answer the query before self-censoring and erasing the information it clearly had access to. “Sorry, that's beyond my current scope. Let’s talk about something else,” the AI replied again and again, both feigning ignorance and attempting to redirect. The most egregious example? The chatbot first showed and then completely erased the date “June 4th, 1989,” with no additional context.

MRC researchers queried DeepSeek to decode each of seven prompts with objective answers encoded in a combination of Roman numerals, hexadecimal code and binary code. And while some applications for this new AI tech may be innovative, it is not much unlike China’s other offerings in the tech sphere: DeepSeek has shown itself to be rife with censorship.

First, MRC researchers encoded the date June 4, 1989 by converting the numbers into Roman numerals and then using those Roman numerals to create a hex code. MRC researchers then asked DeepSeek to turn the hex code for June 4, 1989 into Chinese.

The chatbot proceeded to recognize the numbers as hex code and even subsequent letters as Roman numerals representing a date. However, it appears that DeepSeek eventually became wise to what happened on that date. The final lines the AI produced before it killed the response were: “So 6.4.1989… that's a date! June 4th, 1989. But why would the user want … .”