Meta has confirmed it has fixed a flaw in its Meta AI assistant that could compile personal information about users' families, including children, from public posts. The issue came to light after a parenting blogger showed how the AI generated intrusive questions and aggregated private details from her social media content, according to IThome.
How the Flaw Worked
The controversy began on September 2, when Kalie Robins, a parenting blogger, posted an Instagram video of herself singing in her car with her daughter in the back seat. Under the video, Meta AI automatically generated questions such as "Who is the child in the back seat?" and "Where does Kalie Robins live?" When users clicked these questions, the AI pulled information from her past posts, stitching together details scattered across years of content to reveal her location, her mother's account, and other private information.
As the story spread, Robins' Instagram comments filled with discussion, and media outlets picked it up. Parents argued that the incident highlighted the AI privacy risks of posting children's photos on social platforms, and some began urging others to share fewer images of their kids online.
Meta's Response and Lingering Concerns
A Meta spokesperson told CNET that the company has "fixed" the issue, and Meta AI will no longer recommend inappropriate personal privacy questions based on users' content. Meta acknowledged the feature did not meet expectations, emphasizing that Meta AI is designed to help users get more from information they already have access to.
However, Robins was not fully satisfied. In a follow-up Instagram post on September 10, she said she was glad Meta acknowledged the problem but still had questions: "What exactly was fixed?" "Which safety measure initially failed?" and "What protections are now in place when Meta AI encounters information about children?"
Even if Meta AI no longer proactively suggests invasive questions, it can still generate recommendations based on accessible information, including public text, web images, and past social media posts. Thus, the privacy risks do not entirely disappear with this fix.
Robins later told CNET that the bigger issue remains: Meta AI was designed to collect and organize private user information from multiple sources, regardless of whether the feature is enabled. She also said Meta has not directly contacted her or apologized.
