Meta says it’s changing AI suggestions after posing invasive personal questions

Meta announced it is changing the suggested prompts shown by its AI assistant after an Instagram user posted a viral clip alleging the feature probed for personal details about her children. The post, shared by Kalie Robins, displayed a suggested prompt under a cross-posted Facebook video that asked "Who is the child passenger?". When Robins followed the prompt, she says the assistant compiled information about her daughters from earlier posts and relatives’ accounts and offered further prompts about their ages and her family’s location.

In a statement to The Verge, Meta acknowledged the suggestion was inappropriate and said the company had "missed the mark," adding the assistant should not have presented such questions. Meta also said it has fixed the underlying problem that caused prompts to steer toward personal topics. The company noted that when a user selects a prompt and requests information, the assistant’s replies are based on material the user already has access to, and in this case it may have drawn from videos where the children were named.

The incident adds to recent scrutiny of Meta’s AI features. The company has integrated its assistant across Facebook, Instagram, WhatsApp and Messenger to help with tasks such as answering questions, drafting posts and generating images. In July, Meta removed a separate capability that let people create AI-generated deepfakes of other Instagram users after public backlash.

Meta’s response follows earlier reporting of the episode by Futurism and the viral sharing of Robins’s video on Instagram. The company says it has corrected the issue that led to the invasive suggestions; it did not, in the information reported, provide a detailed technical explanation of the cause or describe any additional safeguards it plans to deploy.