France Halts Chinese AI Model Over 'Bias' Concerns

France's Ministry of the Economy and Finance, known as Bercy, has suspended the use of an artificial intelligence model developed by Chinese company Alibaba following alerts from senior officials regarding "Chinese bias" in its responses. The AI tool, named "HéphAIstos," was being piloted by approximately 100 agents within the prestigious Directorate General of the Treasury (DGT) since early June.

“

responses oriented or biased on subjects relating to China.

The DGT is responsible for monitoring French and international economic conditions, developing growth and employment forecasts, and advising the government on various policy areas, including international trade and taxation. The "HéphAIstos" tool included a conversational agent designed to assist high-ranking civil servants with daily tasks, as well as a multilingual transcription function. However, some users reported "oriented" or "biased" answers, particularly on subjects concerning China.

Concerns about the AI's neutrality were amplified by a high-ranking official, who expressed that the choice of a Chinese-developed model was "extremely serious" for advising the French government on its trade policy with China. This official suggested that a software developed under the watchful eye of Chinese authorities would "necessarily have a pro-Chinese bias."

“

extremely serious for the way in which the French government is advised on China's trade policy because there is necessarily a pro-Chinese bias with software developed under the watchful eye of the authorities.

Bercy confirmed the immediate halt to the experiment due to observed biases in responses related to Chinese subjects. A new model from the French startup Mistral AI has since been installed. While Bercy did not provide specific examples of the biases encountered, Chinese chatbots have previously faced scrutiny for providing laconic or misleading information on sensitive topics like the repression of Uyghurs and Taiwan's political status. Alibaba Cloud, the AI division of the Chinese tech giant, declined to comment on the matter.

“

interrupt the experimentation immediately after only a few days due to alerts concerning biases observed in certain responses related to Chinese subjects.