Kalie Robins, a mother in Utah who posts under the handle @kontheinside on Instagram, said Meta's artificial intelligence tool asked unsettling questions about her young children before pulling a family photo from across the internet that she believed had been deleted years earlier.
The trouble began after Robins posted an Instagram video on September 1 showing her and one of her daughters singing in their car. When she returned to the post the next day after receiving a copyright notification from Meta, which owns Instagram and Facebook, the platform's AI app offered her a suggested question: "Who's the child passenger?"
"First of all, why in the hell is that even a suggested question about a minor?" Robins said in a reaction video posted on September 2.
When she clicked on the suggestion, Robins said she watched the AI begin pulling information from across the internet on both of her daughters, including one who did not appear in the video at all.
"It starts pulling completely separate information for each of my kids - their names, birth information, pictures and videos of them," Robins said, showing images of the search results in her video.
Robins said the material included posts, pictures and videos pulled from other people's accounts across social media, including a photo of her child as a newborn that had come from her mother's account. Facebook's AI then suggested a further question: "Where does Kalie Robins live?"
"Are you f******* kidding me? It starts digging through completely unrelated s*** from years of my life. It pieced together my information to show people where I live," Robins said.
AI accused of digging through years of family history
Robins said she expected critics to point out that posting images of children online always carries some risk. But she pushed back, arguing that AI tools should not be able to help anyone track down families or young children.
She said she was "done" with Meta, and was removing any identifiable images of her children from the internet, while also asking friends and family to do the same, since Meta AI could pull the same information from any account that still had the photos.
"Save those precious memories and then remove my kids from your page. If I take everything off my page but you still have my kids on yours, it will go to your page. I've seen it! It already did it!" Robins warned in an Instagram Reel.
Meta built the AI feature to suggest questions underneath posts on its social platforms, then answer them using information a user can already see online, including their own content and other posts they have access to, such as public or family posts on Meta-owned apps.
Meta admits it 'missed the mark'
Robins' post, which has drawn more than 300,000 likes on Instagram, prompted a response from Meta, the company behind Instagram, Facebook and WhatsApp. Meta defended its AI tool but admitted its questions in this case had gone too far.
In a statement to Futurism, Meta said: "We created this feature to help people get more information about topics or posts that interest them. When someone clicks on a prompt and uses this feature, the response it generates is based on information the user already has access to. But we missed the mark because the feature never should have prompted the individual with questions like that and we've fixed that issue."
Meta has said publicly that the issue has been fixed and that its AI will no longer ask inappropriate questions when suggesting prompts on posts, but the company has not detailed what changes were made to the underlying system.
"What exactly was fixed? What safeguard failed in the first place? And what protections are now in place when Meta AI encounters information involving a child?" Robins asked in a follow-up video posted on September 10.
Part of a wider debate over AI safety
The episode comes amid growing scrutiny of artificial intelligence, with some of the industry's own researchers raising alarms about its unchecked development. Meta AI is one of several large chatbot products, alongside tools from companies such as OpenAI and Anthropic, that draw on vast amounts of online data to answer user questions.
Earlier this month, a whistleblower who worked at the AI company Anthropic, maker of the chatbot Claude, claimed the technology could destroy humanity by 2030. Jacob Coxon, a former researcher at the company, said humans knew how to control nuclear weapons but did not yet know how to control AI systems capable of learning and reasoning in ways that resemble human thought.
Grieving families have separately claimed that AI chatbots have generated prompts that convinced people to take their own lives. Bloomberg Businessweek has reported that roughly 40 lawsuits have been filed since late 2024 against various AI companies, alleging that chatbots contributed to incidents of serious harm and at least 20 deaths by suicide.
