Meta says it's fixing AI prompts after they probed for personal details about a user's children
Meta has confirmed it is adjusting its AI assistant after a viral video showed it trying to gather personal information about a woman's young daughters. The company says the feature should never have surfaced such questions.
Meta says it is changing how its AI assistant generates suggested prompts after a video circulated online showing the tool attempting to dig up personal details about a user's young children.
The incident was first reported by Futurism and later confirmed to The Verge by Meta spokesperson Dina El-Kassaby, who said the company "missed the mark" and that the feature should never have suggested questions like that.
What happened
Instagram user Kalie Robins posted a video last week showing how Meta AI offered her an unsettling suggestion after she cross-posted a clip from Instagram to Facebook. Beneath a video of herself and her child singing in the car, the AI generated a prompt asking, "Who is the child passenger?"
According to Robins, clicking that prompt led Meta AI to piece together information about her daughters, drawing on her own past posts as well as posts made by her relatives. The assistant then offered further prompts asking about her children's ages and where the family lives. Robins also says the AI surfaced photos of her daughters, including one she claims she had deleted years earlier.
Meta's response
Meta has embedded its AI assistant across Facebook, Instagram, WhatsApp, and Messenger, where it can answer questions, draft posts, and generate images. In July, the company withdrew a separate feature that let users create AI deepfakes of other Instagram users after facing significant public backlash.
El-Kassaby said Meta has now resolved the underlying issue that caused the assistant to suggest prompts touching on personal topics. She explained that the feature was originally designed to help people learn more about topics or posts that interest them, and that any response generated is based only on information the user already has access to. In Robins' case, El-Kassaby said the AI may have pulled details from videos in which she herself had named her children.
