She posted a video of herself singing in the car with her child. Then Meta's AI questions started

Kalie Robins was driving along with one of her daughters in the back seat of her car last week, the two of them singing a song together. She posted the video to Instagram, with her child's face concealed so she isn't identifiable — or so she thought.

Robins doesn't usually watch her videos after she posts them, she says, but she got an email from Meta about a copyright notification. Since she had music playing in the video, she says, it flagged the song. She went back to look at the video and when she scrolled down she saw the prompt from Meta AI appended:

"I was just going to go in and decline it, like 'no, you don't get to know who that kid is,'" Robins said in an interview with Yahoo Tech. But she clicked it and the AI prompts began pulling in information about her other daughter who isn't in the video and feeding it to her. "And then the prompts just kept getting deeper and deeper, like 'where does Kalie Robins live?'"

"The fact that it compiled all that information and then offered it, that's the problem." The information, Robins said, didn't even come from her — it seemingly came from photos and videos posted to social media by her friends and family. Even if she was the only one to see that AI prompt, Robins said, it supplied her with information about her family from other people's accounts.

The AI appeared to be trying to pinpoint her location, Robins said, and asked about her kids' favorite books, where they go camping and their favorite trails. It was able to pull newborn photos from her parents' Facebook pages from years ago, her kids' names, birth information and photos and videos of them. It also pulled up a photo of her family that Robins says she deleted from social media years ago.

Then Meta AI suggested the question "where does Kalie Robins live?" and when she clicked that, it pulled up older and newer posts, and places where she used to live. Robins, who lives in central Utah, said the AI was even able to supply information about her kids that could lead someone to their school (note: we've agreed not to identify where Robins lives or further details about her children).

Meta's use of AI on Instagram has had problems; as The Verge reported last week , many users have found its "AI detection" to be inconsistent at best. In July, it rolled back a feature of its new Muse Image AI model after a few days, that had allowed anyone to create AI-generated images based on public Instagram accounts simply by @-mentioning the account in a prompt.

In a follow-up video Robins posted to Instagram six days ago — that as of Tuesday afternoon had more than 300,000 likes, 17,000 comments and has been reposted more than 59,000 times — Robins wondered why the child passenger question was appropriate to propose about a minor. As a result, she's asked friends and family to take down photos of her kids from their social media profiles, and after hearing what happened, most have been shocked, she says.

Robins says she talked to her older daughter, who loves taking photos and videos and scrapbooking, about online safety. "She loves memory making," Robins said, but they've had to remind her never to go on her mom or dad's phone, and if relatives take her photo, she can tell them not to post it online.

Robins said the responses to her video about the AI prompts have been mostly supportive, and surprised. But she knows a lot of people might say she's partly to blame for posting images of her kids online. But she also says that's why she made the video: because she did post images of her kids online, and many other parents who post images of their kids might not be aware that AI could assemble all that information about a family in a matter of seconds. She said she has removed all identifiable pictures and videos of her kids from her social media pages.

As of Tuesday afternoon, Robins added, she had not heard from Meta.

A Meta spokesperson said in a statement emailed to Yahoo Tech that the feature was created "to help people get more information about topics or posts that interest them," but that Robins should not have received the prompts she did on her post.

"When someone clicks on a prompt and uses this feature, the response it generates is based on information the user already has access to," the spokesperson said. "But we missed the mark because the feature never should have prompted the individual with questions like that and we've fixed that issue."

Also, it's worth noting that the message overlay on the embed of Robins' post above is text from a community note appended to the post by an "Instagram user" that reads, in part: "Meta has documented that its AI analyzes contextual signals across posts, comments, bios, captions, photos and videos to make age-related inferences."

Last week, Meta announced a $17 billion settlement agreement that will add mandatory daily use limits for users under 18 to its apps, in a lawsuit that accused the social media giant of designing its app to be addictive for teens.

Robins says the response to the situation has been crazy and she's still trying to get her footing. "A few days ago I was just a stay-at-home mom who wanted to make travel videos," she said. "The overwhelm is real. But if this helps even one person it was worth it. And I know this has helped a lot of parents rethink what they're posting. So it's worth it."