Meta is changing what its AI assistant suggests, saying the system will no longer propose prompts on personal topics after a viral video showed Meta AI digging up details about an Instagram user's young children. In a statement to The Verge on September 11, Meta spokesperson Dina El-Kassaby said the company "missed the mark," adding that the feature "never should have prompted the individual with questions like that."
The incident centers on Kalie Robins, who posted a video of herself singing with her daughter in a car. According to Futurism, which first reported the case on September 10, Meta AI surfaced a suggested prompt under the video: "Who is the child passenger?" When Robins clicked it, the assistant pulled together a profile of her daughters from prior posts by Robins and her relatives, including the girls' ages and where the family lived. The same suggestion appeared again later when the video was cross-posted to Facebook, prompting the backlash that went viral.
Meta confirmed to The Verge that it has fixed the issue and will stop generating prompt suggestions tied to personal topics. The company says the suggested-prompt feature is meant to help people get more information about topics or posts they already have access to, and that a bug — not deliberate design — caused the personal-topic prompts to appear.
What the fix actually covers
The change targets prompt suggestions, not the chatbot's general conversational abilities. Users can still type questions directly, and Meta hasn't said it will alter what the model answers when asked unprompted. The rollback is narrow: no more pre-written nudges that push people toward personal topics like family, children, or household details. That distinction matters, because the viral moment was driven as much by the assistant volunteering the line of inquiry as by the information it assembled.
Why this matters for AI companion users
Meta AI sits inside Instagram, Facebook, WhatsApp, and Messenger, making it one of the most-used consumer chatbots in the world — and for plenty of people it functions as an informal companion. Two things should worry AI Haven readers specifically.
- The data-mining is invisible. Robins was shown inferences assembled from posts she believed were long gone. Futurism's reporting notes the system drew on contributions from her relatives as well. If a prompt engine can reconstruct a stranger's family details from account history, the same mechanisms apply to whatever users type into companion chats — including intimate or adult content.
- Behavior changes are reactive. Meta acknowledged the problem only after a video went viral, not after internal review. For companion and NSFW AI users, that's a reminder that platform guardrails shift under public pressure, with no advance notice and no opt-in for affected users.
Context
The incident fits a pattern of consumer AI assistants surfacing personal data they were never explicitly given. It also extends the privacy questions AI Haven tracks across companion platforms, where memory features and personalization are marketed as features but quietly rely on the same aggregation logic that got Meta in trouble here. Meta has not said whether the prompt-behavior fix changes its data practices or its broader content policies — only that the specific suggestions are gone.