Meta says it fixed AI prompts after invasive questions about kids
Instagram user Kalie Robins posted a video, first reported by Futurism, showing that Meta AI had suggested an invasive prompt after she cross-posted a clip to Facebook. The clip showed Robins and her child singing in the car, and Meta AI displayed a suggested question beneath it: "Who is the child passenger?" When Robins selected the prompt, she says Meta AI pieced together information about her daughters by drawing on her own earlier posts and posts from her relatives, then offered further suggested prompts asking about the children's ages and where the family lives. Robins says Meta AI also surfaced photos of her daughters from the platform, including one she says she had deleted years earlier.
Meta spokesperson Dina El-Kassaby told The Verge that the company "missed the mark," adding that "the feature never should have prompted the individual with questions like that." She said the underlying issue that caused Meta AI to suggest prompts about personal topics has now been fixed. El-Kassaby said Meta built the feature "to help people get more information about topics or posts that interest them," and that the assistant can only surface content the querying user already has access to: "Meta AI returns responses containing content that the querying user can already access. If the user can't see a post, then Meta AI won't return it." In this case, according to El-Kassaby, Meta AI may have pulled from videos where Robins had named her children herself.
Meta's AI assistant is built into Facebook, Instagram, WhatsApp, and Messenger, where it can answer questions, write posts, and generate images. Separately, in July, the company had already pulled a different feature that let people create AI deepfakes of other Instagram users, after facing fierce backlash.
Key facts
- Instagram user Kalie Robins found that Meta AI suggested the prompt "Who is the child passenger?" under a video of her and her child singing in the car, after she cross-posted the clip from Instagram to Facebook.
- When Robins selected the prompt, Meta AI pieced together details about her daughters from her own past posts and from relatives' posts, then suggested further prompts about the children's ages and where the family lives.
- Meta AI also surfaced photos of her daughters, including one Robins says she had deleted years earlier.
- Meta spokesperson Dina El-Kassaby told The Verge the company "missed the mark" and said Meta has now fixed the issue that caused the AI to suggest prompts about personal topics.
- El-Kassaby said Meta AI can only return content the querying user already has access to, and said the assistant may have pulled from videos in which Robins had named her children herself.
Why it matters
Meta AI's integration with Facebook and Instagram is built to draw on a user's own content, including material posted by relatives, to answer questions and generate suggestions. This case shows what that design can produce at its most uncomfortable: technically permitted access to old photos and captions from multiple accounts, assembled into unsolicited follow-up questions about a specific child's age and where the family lives. Meta's public admission that the feature "missed the mark," rather than a defense that the output was merely technically accurate, signals the company sees a gap between what an AI assistant is allowed to access and what it should proactively surface, particularly around children.
Who it affects
Anyone using Facebook, Instagram, WhatsApp, or Messenger, since Meta's AI assistant is built into all four; more specifically, parents like Kalie Robins who post photos and videos of their children, and their relatives, whose own posts about the same children fed into the AI's suggestions. It also affects users generally who assumed a deleted post was gone: Robins says one of the photos Meta AI surfaced was one she had deleted years earlier.
How to use it
Meta AI's suggested prompts appear under a user's own photos and videos, offering to answer questions about what is in them; the company says it built the feature to help people learn more about topics or posts that interest them. El-Kassaby says Meta has already fixed the specific issue that caused the assistant to suggest prompts about personal topics such as a child's identity, age, or location. The assistant remains active across Facebook, Instagram, WhatsApp, and Messenger, where it also answers questions, writes posts, and generates images unrelated to this issue.
How solid is it
The account rests on two sources: Kalie Robins's description of what she saw, relayed through her video and paraphrased by the article rather than quoted in her own words, and an on-the-record statement from a named Meta spokesperson to The Verge. Meta's own explanation of the mechanism is hedged: El-Kassaby says the assistant "may have" pulled from videos where Robins named her children, not that it definitely did. Meta's broader claim, that the assistant only surfaces content the querying user can already see, is the company's own description of its system rather than an independently verified fact. The Verge later updated the article, on September 11, adding further detail from Meta on what information its AI responses can access.
Risks and caveats
The source does not say when the fix rolled out, and Meta's own account of the mechanism is a hedge, not a confirmed technical explanation, so nothing here independently verifies the issue is fully resolved. No regulatory, legal, or child-advocacy response to the incident is mentioned. The daughters' ages, a view or share count for the viral video, and a direct quote from Robins in her own words are all absent; her account is entirely paraphrased by the article. Meta separately pulled an Instagram AI deepfake feature in July after backlash, but the source does not tie that removal causally to this prompt-suggestion issue; they are reported as two distinct problems with Meta AI features.
“the feature never should have prompted the individual with questions like that.”
— Dina El-Kassaby, Meta spokesperson