Meta AI suggested questions about a user's children and then assembled details about them from years of Facebook posts, according to an account from a mother of two that Malwarebytes Labs published today. Kalie Robins posted a video of herself and her young daughter singing in a car in early September, and the system responded with a suggested prompt asking who the child passenger was. Clicking it, she said, produced names, birth dates and videos for each of her kids.
The results went beyond what she had posted herself. The system surfaced a photo of her newborn daughter that Robins' mother had uploaded years earlier, along with a picture Robins said had long been deleted from her own account. A second suggested question asked where Kalie Robins lives, and she said the system worked through old posts hinting at past addresses and newer ones tying her to her current home before trying to pinpoint her location.
Meta said the prompts should not have appeared. "The feature never should have prompted the individual with questions like that," the company said in a statement to the Verge, adding that it had missed the mark and that the underlying issue had been fixed. A spokesperson also said the AI only surfaced information the person asking could already access, an explanation that does not account for the deleted photo Robins described.
Meta AI launched in April 2024 and now runs inside Facebook, Instagram, WhatsApp and Messenger, where it answers questions, generates images and helps people create or look up information. The episode fits a pattern the publication traces back through earlier incidents, including users publicly sharing their Meta AI conversations without realizing it and a bug that could have exposed private chats to anyone guessing user IDs.
Robins said she is done posting identifiable images of her children and plans to remove what is already up, and to ask friends to do the same, on the reasoning that Facebook could otherwise harvest her kids' pictures from their accounts. The report's own advice is blunter: the clearest protection is not posting identifiable photos of children, whether they are yours or someone else's.
The incident lands amid mounting scrutiny of how Meta's platforms treat minors. A coalition of states is in federal court in Oakland, California, over allegations that Facebook and Instagram features were knowingly designed to hook children, with four states trying the case first and 25 others expected to follow. Those claims are being litigated and have not been established; Meta disputes them and says its record will show a commitment to supporting young people.
Regulators in Europe have moved on similar ground. The EU Commission has said a Digital Fairness Act expected by the end of 2026 will target addictive design practices at TikTok, Meta and X and impose strict limits on how platforms use AI. What the Meta AI episode does not include is any figure for how many accounts saw the child-related prompts, or any detail on how long the behavior persisted before the company says it fixed it.
