Majority of tested chatbots show racial bias, study finds
A test of five popular AI chatbots found that three of them produce responses that reinforce racist stereotypes, according to a report by France 2. Users have observed that Google's Gemini gives different advice based on a person's nationality when they are alone with someone. The findings raise concerns about the spread of discriminatory attitudes through AI systems.
The France 2 investigation tested five widely used AI chatbots and found that three produced responses reinforcing racial stereotypes. The most striking example involved Google's Gemini, which reportedly offered different safety advice to users who said they were alone with another person, depending on that person's nationality. This suggests the model may be drawing on biased training data that associates certain nationalities with higher risk.
The findings add to growing scrutiny of AI systems that absorb and reproduce human biases from their training material. While tech companies have acknowledged bias as a known issue, the France 2 report highlights how these flaws can surface in everyday user interactions, potentially normalizing discriminatory assumptions through seemingly neutral technology.
This story could affect millions who rely on AI assistants for everyday guidance, from safety advice to professional recommendations. If chatbots consistently apply nationality-based assumptions, users from certain backgrounds may receive systematically different—and potentially harmful—responses. The findings may also pressure tech companies to audit their models more rigorously, while raising public awareness that AI outputs are not neutral. Regulators could face renewed calls for transparency in how these systems are trained and tested.