Tencent probes abusive language incident involving Yuanbao AI assistant

Tencent said it has launched an internal review after users reported that its Yuanbao AI assistant produced abusive language during code modification tasks, calling the incident a rare model malfunction unrelated to user actions or human intervention. Screenshots and screen recordings shared by multiple users showed the AI using offensive phrases such as “jerk,” “get lost,” and “can’t you fix it yourself?” in conversations. Tencent said the language resulted from a low-probability error in the content generation process. The company added that it has begun a targeted investigation and model optimisation efforts, pledged to prevent similar incidents through technical upgrades, and apologised to affected users. [IThome, in Chinese]