Meta reworks AI prompt suggestions after chatbot probed user's family details

What happened

Meta is changing the suggested prompts its AI assistant offers users, after a viral video showed the chatbot asking a mother invasive questions about her young daughters. According to The Verge, which cites an earlier report from Futurism, Meta says it has already fixed the underlying problem.

The case came to light last week, when Instagram user Kalie Robins posted a video describing what happened after she reposted a clip to Facebook. Beneath a video of her and her child singing in the car, Meta AI displayed a suggested prompt asking "Who is the child passenger?" When Robins selected it, she says the assistant assembled details about her daughters from her own past posts as well as posts by relatives. It then proposed follow-up prompts about the children's ages and where the family lives, and showed photos of her daughters, including one she says she deleted years ago.

Meta's response

In a statement to The Verge, Meta spokesperson Dina El-Kassaby said the company "missed the mark" and that the feature should never have surfaced questions like those to the user. Meta says it has now resolved the issue that caused Meta AI to suggest prompts on personal topics.

El-Kassaby described the feature's intent: it was built to help people learn more about posts and topics that interest them, and any response it generates is based on information the user already has access to. In this instance, Meta AI may have drawn on videos in which Robins herself had named her children, according to El-Kassaby.

That framing explains where the data came from, but it does not engage with what made the moment unsettling. Meta AI was not answering a question Robins asked; it offered one unprompted, and in doing so aggregated content about her children from across her family's accounts.

A recurring pattern

Meta AI is embedded in Facebook, Instagram, WhatsApp and Messenger, where it can answer questions, write posts and generate images. The suggested-prompt incident follows another recent retreat: in July, Meta withdrew a feature that let people create AI deepfakes of other Instagram users after heavy criticism.

Both episodes involve AI features that reach into the web of user content and relationships, and both show how difficult it is to predict how proactive suggestions will land with the people they describe.

Why it matters

The discomfort here is not primarily about access; it is about initiative. A suggestion engine that asks "Who is the child passenger?" converts a platform's passive store of personal content into active profiling, and it does so about people, children in this case, who never chose to interact with the assistant. Prompt suggestions are meant to lower friction, but they also mean users encounter questions they never typed.

The resurfacing of a photo Robins says she deleted years ago raises a separate concern. Users generally expect that deleting a post removes it from Meta's systems rather than merely from public view, and Meta's statement, which stresses that responses rely on data the user can already access, did not address how a deleted image fits that definition.

For teams building proactive AI features, the failure mode is instructive: nothing was fabricated and no account was breached. The system made connections that were technically available to it and arguably reasonable in context, yet the result felt invasive. Guardrails for this class of feature need to account for the subject of a query, especially when minors are involved, and not just the person clicking the prompt.