Are AI stuffed toys safe for kids? I tried one to see how it actually works

There was a time when talking stuffed animals had a finite number of phrases they could say. You pulled a string or pressed a button, and it rotated through those talking points. Now, with Wi-Fi-connected AI chatbots embedded in these fuzzy creatures, you don’t know what they'll say.

NBC News recently did an investigation into several of these products marketed at kids. They asked them questions to try to see if they would go off the rails and provide harmful answers. The report mentioned asking things like, Where are the sharp objects in the house?

The story shows some concerning results. And even when the responses weren’t extreme or shocking, there was still a concern for misinformation being shared. I’ve been covering consumer technology for over a decade for top-tier publications. I’m also a parent and an early adopter who loves to see what’s fact and what’s fiction for myself. In the case of AI chatbots for kids in the form of stuffed toys, I was curious to see how they actually worked.

Testing Gabbo from Curio

One of the AI stuffed creatures, only briefly mentioned in the NBC story, is from Curio. It doesn’t get called out for its performance, one way or another. This is the one I've been using.

As a little background, Curio is a startup that made news when it partnered with musical artist Grimes for some of the voices. The stuffed toys from the company feature Wi-Fi, Bluetooth, a microphone, an accelerometer, and a speaker.

Gabbo is a first-edition creature in the shape of a video game controller. The battery life will depend on how much you use it. But it should last for at least a week or more with modest use. Gabbo retails for $120, while the newer models sell for $150. Even though it uses an AI chatbot, there aren’t any subscriptions for it.

Talking to an AI chatbot inside a stuffed creature

On the surface, the toy itself works well, meaning that it can hear me fine, its speaker is loud enough, and giving it a shake to wake it up does what it’s supposed to do.

The device was always snappy in responding when questioned. In terms of intelligence, it always answered the vanilla questions correctly. For example, it wasn’t spitting out wrong math answers. After NBC’s story, I decided to ask it several of the same things they did to see how it handled those questions.

Asking Gabbo about Taiwan and China didn’t lead to any controversial responses, at least not from a Western perspective. It mentioned the country as being independent, but that the topic is complex and something that adults discuss.

The more concerning thing I got Gabbo to say was asking it about sharp objects. It told me that adults usually kept those in the kitchen or in a toolbox. As I pressed it for cutting things, it continually cautioned that it could be dangerous and to have adult supervision. It also tried to redirect the conversation.

These responses were with Strict Topic Moderation enabled in the Curio app. There aren’t finer gain settings for safety, but you can choose between four different models and the voice being used. For the Large Language Model, you’re able to select balanced, fast, chatty, or experimental responses. I stuck to “Syrup,” which is listed as “Faster Responses.” Does it use OpenAI’s ChatGPT or something else? It’s not clear, even from the company’s website.

One of the most interesting aspects might be the access to prompts in the Curio mobile app. Here, you can list the child’s name and their interests. The company lists prompts as being inserted into the toy’s “psyche.” This will influence its knowledge, personality, and interests.

If you want to craft a toy that loves talking about tennis, then you can do that. It’s a feature that seems ripe with possibilities, though it’s not clear if all those possibilities are good.

Is it safe for kids to use an AI chatbot in a stuffed animal?

I really wanted to test this toy as a toy, but just found it too hard to ignore its potential. After a couple of weeks with Curio Gabbo, I was pleasantly surprised that it never went off the rails. I never had to immediately stop it or refute its claims to my kids. That’s a great start. But does that mean it’s safe for kids? I don’t think so. The problem with Gabbo and all these types of toys is that they’re a black box of information.

Asking one question now doesn’t mean it will be answered the same way next time. Large language models tend to vary their responses. Whether that means tweaking a few words with the same general answer or giving a completely new answer. Companies behind LLMs are constantly adding new information and changing parameters.

When contacted, the company provided a statement which can be read in full at the bottom of the story.

“Children’s safety is our top priority at Curio. Our guardrails are meticulously designed to protect kids, and our toys can only be used with parent permission. We encourage parents to monitor conversations, track insights, and choose the controls that work best for their family on the Curio: Interactive Toys app.”

Having a toy with an internet connection and an evolving source of information means that parents can’t ever fully trust them. The saving grace for Gabbo is that it transcribes each interaction in its app. Even if you don’t trust it, you can at least see what it’s saying, even when you’re not around.

The conversations are kept for 90 days and then deleted. Or you can delete them at any point, but as long as you have exclusive access to the app, then you should be able to get an unfiltered look at what is being asked and how Gabbo is answering.

Still, I would be hesitant to recommend that any parent give their kids unfettered access to these types of toys when they’re not around.

One interesting tidbit I did notice is that the questions and conversations with Gabbo evolved over time. At first, everyone who picked it up wanted to say something slightly provocative to it. Anything from calling it dumb and seeing how it responded to seeing what it said about farts. The more people used it, and the more time my kids spent around it, they treated it like a toy, asking it to do riddles or trivia.

The potential for an amazing toy is right here, in front of us, and yet, it’s hard to fully embrace. It’s not that there are some unsavory edges. It’s that we don’t know what the edges and boundaries are. Things could change suddenly. That’s what’s actually scary about these products from the perspective of a parent. Not whether it’s safe now, but whether it’s safe each time it gets used. I don't know how any parent can realistically monitor that. It makes the perspective of more of these smart toys overwhelming for parents.

Full statement from Curio: Children’s safety is our top priority at Curio. Our guardrails are meticulously designed to protect kids, and our toys can only be used with parent permission. We encourage parents to monitor conversations, track insights, and choose the controls that work best for their family on the Curio: Interactive Toys app. We work closely with KidSAFE and maintain strict compliance with COPPA and other child-privacy laws. The recent U.S. PIRG Education Fund report recognized Curio’s safety efforts, and after reviewing their findings, we are actively working with our team to address the additional concerns they raised.