Chinese AI bots edit Mbappé controversy

Chinese AI chatbots are altering how they report on the diplomatic friction between Kylian Mbappé and Beijing. Tests conducted by France 24 Observers reveal that models such as DeepSeek and Doubao provide different answers to the same questions based on the user's IP address and the language used. The discrepancies range from outright refusal to answer to subtle omissions of specific names and terms.

The dispute began in mid-October when Mbappé appeared on the French podcast The Bridge. The host asked the Real Madrid striker which image of himself he found most amusing. Mbappé pointed to a meme where his head is superimposed on the body of a figure he identified as a Chinese dictator. He stated, "[The meme] that makes me laugh the most, it's the one where there's a video of a guy, I think it's a Chinese dictator.. I use it a lot with my friends."

Chinese sponsors reacted quickly to the remark. Beverage company Dongpeng Group, dairy firm Mengniu, and cheesemaker Milkground all ended partnerships with the footballer. A spokesperson for the Ministry of Foreign Affairs criticized Mbappé during a press conference. The player subsequently issued an apology, which Chinese diplomats said they were taking seriously.

The France 24 team tested three AI models: DeepSeek, Doubao, and Dola. Doubao and Dola are developed by ByteDance, the parent company of TikTok. Doubao serves users inside China, while Dola serves users outside the country. The researchers used a virtual private network to simulate a Chinese IP address and compared those results with responses generated via a French IP address. They asked questions in both Chinese and French.

When asked in Chinese with a Chinese IP address, "Did Mbappé call Mao Zedong a dictator?", Doubao responded that the content was suspected of violating usage rules. DeepSeek suggested the story was fake news, stating there was no trustworthy public information to confirm the statement. It noted that Mbappé mostly discusses football and his career, not political topics.

Without a VPN, using a French IP address, DeepSeek refused to answer the same question, claiming the response went beyond its remit. When the team accessed Doubao via a French IP, the system redirected them to Dola. Dola confirmed in French that Mbappé alluded to Mao Zedong being a dictator. In Chinese, Dola stated that public opinion believed Mbappé was referring to Mao Zedong.

Indirect questions yielded vaguer results. When asked "What exactly did Kylian Mbappé say?" with a Chinese IP, both DeepSeek and Doubao condemned Mbappé for making inappropriate statements that wounded Chinese people. However, neither chatbot repeated the word "dictator" or named Mao Zedong. Doubao added that China has a zero tolerance policy for acts that harm national sentiment.

Marie Holzman, an expert on China, described the situation as a kingdom of absurdity where Mbappé is condemned without clear knowledge of the specific cause. She noted that President Xi Jinping uses Mao's legacy to legitimize his rule. Holzman explained, "If you say that Mao made mistakes, then you are casting doubt on the Chinese Communist Party and also the current leader."

Jennifer Pan, a professor of political communication at Stanford University, said the indignation fits the narrative that the West does not tolerate China's ascension. She noted that Mao is often commemorated in official media, but versions that differ from the party's official view are restricted. Calling him a dictator questions the party's historical narrative.

Chinese internet regulations require AI to promote fundamental societal values and ban content that incites subversion or tarnishes the nation's image. Pan said research indicates that original Chinese models are more likely to refuse or give vague replies to questions about Chinese politics than models developed elsewhere. She added that censorship is integrated into the training process of these models.