Meta says it’s changing AI suggestions after posing invasive personal questions

Details

Meta's AI assistant surfaced a deeply invasive prompt to Instagram user Kalie Robins when she cross-posted a car video to Facebook: "Who is the child passenger?" When she engaged, Meta AI compiled personal details about her daughters by cross-referencing posts from her own account and relatives' accounts — offering follow-up prompts about the children's ages and her home location, and surfacing photos she believes she deleted years ago. Meta spokesperson Dina El-Kassaby acknowledged the company "missed the mark" and said the feature "never should have prompted the individual with questions like that." The company's explanation — that responses are based on information the user already has access to — sidesteps the core issue: aggregating dispersed data across a social graph can produce a detailed family profile that no single post would reveal. This is not the first time Meta AI has overreached. In July 2026, Meta pulled a feature allowing AI deepfakes of other Instagram users after user backlash. The pattern — launch to billions, face public outrage, roll back — is becoming familiar for features deployed across Facebook, Instagram, WhatsApp, and Messenger. The fix announced is narrow: personal-topic prompt suggestions are disabled. Meta has not described any change to how its AI indexes or aggregates data from across accounts, meaning the underlying capability that enabled this behavior remains in place.