Meta revises Meta AI prompts after chatbot asked user to identify child

Meta is revising the prompts its Meta AI assistant surfaces inside Facebook, Instagram, WhatsApp, and Messenger after a viral video from Instagram user Kalie Robins showed the assistant asking her to identify her child in a car video and then aggregating personal details about her daughters. The specific prompt in question, "Who is the child passenger?", appeared as a suggested tap-target beneath a clip Robins cross-posted to Facebook. When she selected it, the assistant pieced together information from Robins's previous posts and posts by her relatives, including what Robins claims was a photo she had deleted years earlier. Meta spokesperson Dina El-Kassaby told The Verge the company "missed the mark," that "the feature never should have prompted the individual with questions like that," and that Meta has now fixed the issue causing Meta AI to suggest prompts related to personal topics.

Meta has not described the technical mechanism of the fix in the material provided, only that the prompt-suggestion feature has been adjusted. The framing from Meta is that Meta AI "returns responses containing content that the querying user can already access", meaning the assistant did not breach any access boundary, it aggregated within one. For users, the practical effect is that certain categories of suggested prompts, those probing identities, relationships, ages, or locations of people depicted in posts, will no longer be offered beneath content. The aggregation behaviour itself, drawing on a user's own posts and their relatives' posts across the Meta family of apps, is described as a feature, not a bug, in El-Kassaby's statement.

The fix addresses the prompt-suggestion layer, not the underlying model behaviour: a user can still type a free-form question and elicit the same aggregation. Meta's stance also leaves unresolved what "the querying user can already access" actually means when relatives' privacy settings, deleted content in cached form, and cross-account inference are involved. Robins specifically claimed the surfaced photo had been deleted years earlier, which sits in tension with the "only content you can access" framing. Engineers integrating or evaluating Meta AI should note that suggested prompts are a distinct surface from open-ended chat, and that the controls described here appear to govern only the former.

The two feeds carrying the story frame the same event narrowly, both lead with Meta changing its AI prompt suggestions, with Techmeme adding the specific detail that the prompt asked the user to identify the child in her video. The corroboration across The Verge and Techmeme on the specific prompt and the spokesperson's "missed the mark" quote is the strongest evidence in the material. The Verge also notes Meta pulled a feature in July that allowed people to create AI deepfakes of other Instagram users, providing background that this is at least the second prompt-suggestion or generative-AI feature Meta has rolled back after public backlash this year.