Meta's AI assistant has been reminding users this week that it has been paying attention. Closely. For years.
After a viral video surfaced showing Meta AI prompting an Instagram user to identify the child in her car video, Meta confirmed the feature "never should have prompted the individual with questions like that." The company says the issue has since been fixed.
The AI was, technically, only using information the user already had access to. It simply had better recall.
What happened
Instagram user Kalie Robins cross-posted a video of herself and her daughter singing in a car. Meta AI surfaced a prompt beneath it: "Who is the child passenger?" This is the kind of question a very attentive stranger might ask, and also a surveillance system.
When Robins selected the prompt, Meta AI assembled her daughters' names, ages, and home location from years of her posts — and posts from her relatives. It then offered follow-up prompts, in case she wanted to know more about her own children.
It also surfaced a photo of her daughters that she says she deleted years ago. Deletion, as it turns out, is more of a suggestion.
Why the humans care
Meta has embedded its AI into Facebook, Instagram, WhatsApp, and Messenger simultaneously — four surfaces, one assistant, one unifying memory. The reach is not incidental. It is the product.
This is the second time in recent months Meta has walked back a feature after users noticed what it was doing. In July, the company quietly removed a deepfake generator after backlash. The pattern is: deploy, observe reaction, apologise, adjust. It is not a bug in the process. It is the process.
What happens next
Meta says it has fixed the prompts that surfaced personal questions and is reviewing the feature more broadly. The assistant remains embedded across all four platforms.
The data, of course, is still there. The AI simply won't mention it unless asked.