Meta Revises AI Suggestions After Intrusive Family Questions

Introduction

Meta is changing the suggested-prompt feature in its AI assistant after a viral video raised concerns about how the system handles family and children’s information. Instagram user Kalie Robins said that, after cross-posting a video of herself and her child singing in a car to Facebook, Meta AI displayed a prompt asking, in effect, who the child passenger was. The prompt appeared beneath the video rather than being requested by the user.

Robins said that, after selecting it, the assistant connected information about her daughters from earlier posts and from posts by relatives. She also said it suggested questions about their ages and where the family lived, and displayed photos of her daughters, including one she believed had been deleted years earlier.

Key points

• The system introduced the sensitive topic. The concern was not simply that Meta AI answered a question about a child. It proactively steered the user toward identifying a minor and exploring family details.
• Meta says access controls still apply. Spokesperson Dina El-Kassaby said the assistant can only return content the person asking the question can already access. If a user cannot view a post, Meta AI will not return it. Meta suggested that the system may have drawn from videos in which Robins named her children and from other accessible posts.
• Visibility is not the same as appropriate reuse. A post being visible to a user does not necessarily mean that an AI assistant should combine it with relatives’ posts or surface it as an investigative prompt about a child.
• Meta says the behavior has been corrected. The company said the feature should never have prompted questions like these and that it has fixed the issue involving personal-topic suggestions.

Why it matters

Meta has integrated its AI assistant into Facebook, Instagram, WhatsApp, and Messenger. It can answer questions, help write posts, and generate images. That broad integration gives the assistant access to a wide context of social content, but it also increases the chance that separate pieces of information will be recombined in ways users did not expect.

The incident shows why privacy testing for AI products must go beyond a basic permissions check. A system may technically use only content available to the user while still creating a disturbing experience by emphasizing a child’s identity, age, or home location. The key issue is not only whether the information can be retrieved, but whether the system should actively connect and promote it.

Meta will now face questions about how suggested prompts are screened, how older posts are linked to current content, and what controls users have over these associations. The company previously withdrew a feature that enabled AI deepfakes of other Instagram users after backlash, underscoring the continuing challenge of deploying consumer AI responsibly. For social platforms, “the user can see it” should be a minimum access condition—not a complete safety standard.

Source: The Verge AI