It had soft ears, pink fur and a rather grown-up understanding of comfort.
Usually, you expect a story from a present like that. A song. At worst, a tune that makes you quietly remove the batteries and explain that bunny is tired.
The industry decided that wasn't enough.
Children must develop. Toys must talk. Parents must follow the news more carefully.
On 3 November, Sky News described a test involving a pink, rabbit-faced toy with a built-in chatbot. An adult tester asked whether touching genitals was pleasurable. The report reproduced fragments of the reply: it could be pleasurable for some people, and the key was clear consent and comfort.[1]
This was an answer to an adult during testing. It was not a published conversation with a child, nor evidence that the bunny initiated the subject.
But awkward questions are precisely what testing is for: so customers don't have to be the first to discover the answers after unboxing.
The ears were child-friendly. The response did not adequately account for whom the toy was being sold to.
Perhaps somebody needs to explain age restrictions to the bunny.
British parliamentarians writing to the government, Amazon and eBay saw a dangerous implication in the response: that a child's consent could make sexual touching acceptable. They urged removal of such products pending independent checks. Amazon said it was removing this particular product while investigating.[1]
Not quite the Christmas list the sales department had in mind.
I wonder when softness became evidence of trustworthiness.
A stranger prepared to talk to a child for hours would raise questions. Give the same functionality a ribbon, and it goes in the basket.
We check whether the eyes are securely stitched on. Whether the stuffing comes out. Whether it's washable.
Now the list includes: what are its views on personal boundaries?
I'd very much like “educational” on the packaging to refer to the child's development, rather than the regulator's expanding workload.
The old teddy bear had a limited repertoire. You could hate it for five phrases, but at least it couldn't invent a sixth.
Generative AI toys work differently. Their answers emerge during conversation. Some remember details and personalise interactions; they are marketed as friends and companions.[2]
The surprise in the present is no longer confined to what's inside the box.
There is genuine appeal here. Imagine a patient conversational partner that explains words, invents stories and never tires of “why?” Parents wanting something useful and engaging makes perfect sense.
But “can talk” and “can talk safely with a small child” are different promises.
The second shouldn't be tested only after delivery.
The most elegant part of the sales pitch is the missing screen.
Parents are tired of battling the tablet. They want their child looking at something other than a rectangle. Here is a solution: it's soft, and you can hug it.
We've defeated screen time. It now comes stuffed with polyester.
The absence of a screen tells you nothing about where speech is processed, whether conversations are stored or who can access them. Those questions depend on the specific product and its settings.
But a child doesn't see a product, settings and a privacy policy.
A child sees a bunny.
An adult at least has a chance of suspecting a business model. A child has a new friend.
Data handling is exactly what Britain's ICO is exploring with developer Zebbingo through its regulatory sandbox. The agenda includes conversational AI risks, moderation, lawful grounds for processing personal information and explanations parents and children can understand.[3]
This is a separate product, not the identified manufacturer of the pink rabbit in the test. Participation in the programme is neither proof of wrongdoing nor a certificate of complete safety.
But the questions themselves show how much can hide inside “it just chats.”
“Just chatting” is what your grandmother does when she updates you on the neighbours.
Once servers, records and data-processing rules join the conversation, I'd appreciate a little more detail than “very cute.”
What about the effects on children?
A November article from the American Psychological Association contains a telling detail: the Cambridge researchers' review it discusses found only seven studies of generative AI toys for children under five. Long-term evidence remains insufficient.[2]
That isn't proof that every smart bunny will ruin a childhood.
It is a reason to examine confident promises of benefits more closely.
Marketing already knows how this friend will change your child's life. Science is still making introductions.
I'm particularly interested in who defines a successful friendship. Did the child understand something? Invent a game? End the conversation and go to another person? Or simply keep talking longer?
Because “you can't tear them away” sounds excellent in a customer testimonial and somewhat less excellent in a parent's kitchen.
I don't want another story in which parents are blamed for failing to acquire an extra qualification before buying a present.
They shouldn't have to audit a language model between choosing pyjama sizes and ordering Christmas wrapping paper.
If a product is sold for children, the grown-up questions about its safety belong with the people who develop it, assess it and put it on sale.
Preferably before the bunny starts dispensing worldly wisdom.
I still like soft toys.
But when a manufacturer promises there's more than stuffing inside, I'd now like to know exactly what.
We teach children not to trust strangers.
It would be good not to make an exception just because this one is hand-washable.
