Meta is changing how its AI assistant suggests prompts after a viral video showed the chatbot asking an Instagram user to identify her young daughter in one of her own videos. The prompt, 'Who is the child passenger?', appeared beneath a clip of Kalie Robins and her child singing in the car after she cross-posted it to Facebook. When Robins tapped the suggestion, Meta AI began stitching together details about her children from her own posts and posts made by her relatives.
Robins says the assistant went further, offering follow-up prompts about her children's ages and where the family lives, and surfacing photos of her daughters from the platform — including one she claims she deleted years ago. The episode, first flagged by Futurism, ran through a feature Meta pitches as a way to help users learn more about topics or posts they're browsing.
Meta spokesperson Dina El-Kassaby told The Verge the company 'missed the mark' with the suggestion.
“the feature never should have prompted the individual with questions like that.”— Dina El-Kassaby, Meta spokesperson
Key facts
- 01Meta AI prompted Instagram user Kalie Robins with the question 'Who is the child passenger?' beneath a video of her and her daughter.
- 02The chatbot then assembled details about Robins's children from her own posts and relatives' posts, and surfaced a photo she says she deleted years ago.
- 03Meta spokesperson Dina El-Kassaby said the feature 'missed the mark' and that the company has fixed the issue causing the personal-topic prompts.
- 04Meta AI is built into Facebook, Instagram, WhatsApp, and Messenger, giving the suggestion engine access to cross-platform user history.
- 05In July, Meta pulled a separate feature that let users create AI deepfakes of other Instagram accounts after public backlash.
El-Kassaby said Meta has fixed the underlying issue that caused Meta AI to generate prompts about personal topics like a user's children, location, or family. She did not detail what the fix changes at the model or ranking layer, or whether the same guardrails now apply across Facebook, Instagram, WhatsApp, and Messenger, where the assistant is embedded.
The company's framing is that the assistant only pulls from information the user already has access to on the platform — in this case, videos where Robins had named her children. That defense sidesteps the harder question, which is why the ranking system decided that a child's identity was a useful prompt to surface at all.
Meta pitches the suggested-prompt feature as helpful curiosity.
“When someone clicks on a prompt and uses this feature, the response it generates is based on information the user already has access to.”— Dina El-Kassaby, Meta spokesperson
The distinction between 'data the user can access' and 'data the user should be nudged to interrogate about a child' is the entire product-safety question, and Meta's statement collapses it. A recommendation engine that can identify a minor from a parent's back catalog, resurface deleted-looking photos, and then invite the parent to ask follow-up questions about that minor's age and location is not a neutral information-retrieval tool. It is a profiling loop with a chat interface on top.
This is not the first time Meta AI has been walked back mid-flight. In July, the company pulled a feature that let users generate AI deepfakes of other Instagram accounts after sustained criticism. The pattern — ship the capability, respond to the viral moment, ship a patch — is now the standard release cadence for consumer-facing generative features across Meta's apps, which collectively reach several billion accounts.
The Robins video also lands in a period of heightened scrutiny of how large platforms handle minors' data in AI systems. Regulators in the EU and several US states have opened inquiries into how consumer AI products treat data belonging to or depicting children, and the FTC has signaled that inferences drawn by AI systems about protected categories can themselves trigger enforcement, regardless of whether the underlying data was technically public.
Meta has not said whether it retrained, re-ranked, or simply blocklisted the categories of prompts involved. It also has not said how many users saw similar suggestions before the fix, or whether the surfaced photo Robins says she deleted is still retrievable through the assistant.
For Meta, the strategic bet on Meta AI is that embedding a general-purpose assistant across Facebook, Instagram, WhatsApp, and Messenger turns the company's social graph into a defensible AI moat that OpenAI and Google cannot replicate. Episodes like this one are the cost of that bet — every consumer surface becomes a live test of what the assistant will and won't say about the people in a user's network. The product question Meta now has to answer is not whether the assistant can infer things about a user's family, but whether it should offer to.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
