Meta Revises Its AI Assistant After Intrusive Questions

After a case of intrusive questioning aimed at a mother and her children, Meta plans to adjust the suggestions of its AI assistant. Spokesperson Dina El-Kassaby states that the anomaly has been corrected and describes a feature intended to rely on information already accessible to the user. The assistant, deployed in the group's main applications, had previously led Meta to remove another sensitive function in July.

Meta claims to have corrected the anomaly and highlights the scope of its AI

Dina El-Kassaby indicates that the issue causing Meta's AI to suggest personal topics has been resolved. She explains that the feature was designed to help users obtain more information about subjects or posts that interest them and that when a suggestion is used, the generated response relies on information the user already has access to. In the case mentioned, the AI may have based its suggestions on videos where the children were named. Meta's AI assistant is integrated into Facebook, Instagram, WhatsApp, and Messenger, with capabilities for answering questions, drafting posts, and generating images. In July, Meta also removed a feature that allowed the creation of deepfakes of other Instagram users following a strong backlash.

The triggering incident described by the user

Last week, Instagram user Kalie Robins recounted the appearance of an intrusive suggestion after sharing a clip on Facebook. Under a video where she sings in the car with her child, the AI displayed "Who is the child passenger?" After clicking, she claims the AI aggregated information about her daughters from her past posts and those of her relatives. She also reports receiving questions about her children's ages and where her family lives. According to her, the AI even displayed photos of her daughters previously posted on the platform, including one she says she deleted years earlier. The existence of the video detailing these elements was reported by Futurism.

Meta announces adjustments and admits a mistake

Meta plans to modify the suggestions offered by its AI chatbot following this incident. Its spokesperson Dina El-Kassaby acknowledges that the company "missed the mark" and believes that the feature should never have posed such questions to individuals.