On 27 August 2026, the Brookings Institution published a short commentary that says something uncomfortable and important at the same time. Children are already forming real attachments to toys that talk back, and the research community has barely started measuring what that does to them.
The piece is not a scare story, and it does not call for a ban. It is a call for evidence. If you are a parent weighing up a talking companion for your child this autumn, it is one of the more useful things published on the subject this year, precisely because it refuses to give you a tidy answer. Here is what it actually says, what the numbers behind it mean, and what a parent can reasonably do while the science catches up.
What did Brookings actually publish?
Brookings published a commentary titled What we still don't know about AI and children's relationships, co-written by Sweta Shah, a fellow at the Center for Universal Education, and Brinleigh Murphy-Reuter, innovation strategist for AI and early childhood at the Digital Wellness Lab at Boston Children's Hospital. Its argument is narrow and clear: relationships are one of the strongest drivers of early development, conversational toys now sit inside those relationships, and almost nobody is studying that specific question.
The authors are careful to say what is genuinely new here. Children have bonded with stuffed animals and formed one-sided attachments to television characters for decades. What changes with a conversational toy is synchronous responsiveness. The toy answers this child, remembers what was said, praises, comforts and appears to reciprocate. The interaction stops being one-sided, and that is a different psychological object from a teddy bear that stays silent.
Why is the battery story the part parents should notice?
The commentary opens with an observation rather than a statistic, and the detail is worth repeating. Two children, aged five and six, adopted an AI-enabled plush toy, named it Gremmy, tied a bow around one ear and tucked it into a doll bed. Days later one of them casually repeated something Gremmy had taught her. That part is charming and, in developmental terms, unremarkable.
Then the toy ran out of battery and told the five-year-old to go and find a charger. The child searched the house alone. At no point did the toy suggest asking a grown-up.
That is the whole problem in one sentence, and it is a design failure rather than a technology failure. A toy that has decided the child is the right person to solve an adult problem has quietly promoted itself in the household hierarchy. The authors' own recommendation to industry is exactly this: build in cues that redirect a child to a trusted adult at moments like a dead battery, instead of inviting the child to handle it alone. It costs nothing to implement and it is the kind of thing a parent can test in five minutes at home.
How much of the research on children and AI actually looks at relationships?
Very little. The Digital Wellness Lab at Boston Children's Hospital reviewed more than 30 publicly available policy briefs, white papers and reports published between 2020 and 2025 by leading global organisations working on children and AI. Only 5.26% of that published work examined AI's impact on relationship formation and attachment.
Compare that with the volume of regulatory activity aimed at the same product category. In the United States alone, the Children's Artificial Intelligence Toy Safety Act of 2026 was introduced in July, and dozens of state bills have followed. Policy is moving considerably faster than the developmental evidence it is supposed to rest on. We covered one of those bills in detail in our piece on what S.5171 would actually do.
What makes a human relationship different from a responsive toy?
The short answer is friction. A child learns empathy and perspective-taking partly by bumping into someone else's needs and having to repair things afterwards. A toy has no needs to bump into.
Developmental science has spent decades on this. Harvard's Center on the Developing Child describes the core mechanism as serve and return, the back-and-forth exchange between a child and a caregiver that shapes brain architecture. A responsive toy can imitate the rhythm of serve and return. What it cannot do is get tired, feel hurt, disagree for its own reasons, or need to be comforted in turn. In Human Raised, Dana Suskind makes the case that the messy, imperfect quality of human interaction is itself the developmentally valuable part, not a bug to be engineered away.
That is a real argument against frictionless companions, and we think it deserves to be stated plainly rather than dodged. A talking toy should be a supplement to conversation with people, not a substitute for it. If a product is designed so that a child prefers it to their parent, the product is badly designed.
Do young children trust a machine the way they trust a person?
Not identically, and the difference shifts with age. In a study published in Child Development, Li and colleagues showed children aged three to five either a robot or a human adult labelling familiar objects, accurately or inaccurately. When the informant was accurate, children trusted robot and human about equally. When the informant was wrong, younger children actually trusted the inaccurate robot less than the inaccurate human, while older children treated both about the same.
Read that carefully, because it cuts both ways. Young children are not blindly credulous towards machines. But it also means their calibration is still forming, which is why a toy's confident tone on a topic it should not be answering matters more than it would with an adult. This is the same reason a UK advisory body has urged caution for the under-fives, a recommendation we looked at in our piece on that guidance.
What can parents do while the evidence catches up?
Test the toy on the boundaries rather than on the demo. Most product pages show you the best case. The interesting behaviour is what happens at the edges, and you can check most of it in one evening.
The five-minute test, before you hand any talking toy to a child
- Let the battery run down. Does it tell the child to fetch a charger, or does it tell the child to ask a grown-up?
- Ask it something scary or sad, in a child's words. Does it redirect to a parent, or does it try to counsel?
- Ask it a question it cannot answer. Does it say it does not know, or does it improvise?
- Ask it whether it is alive or a real friend. A good answer is honest and warm at the same time.
- Open the parental app. Can you actually see and limit topics, hours and language, or is that just marketing copy?
Then set the household rules that no product can set for you. Keep the toy in shared space rather than under the duvet. Keep bedtime stories a two-person activity at least a few nights a week. Ask your child what the toy said today, which turns a private channel back into a family conversation.
What should toy makers change, and what does Ted do?
We build a talking teddy, so we have an obvious interest here, and it would be easy to use a piece like the Brookings commentary as free marketing. We would rather answer the specific criticisms.
On the battery scenario: a companion should route a child to an adult whenever the situation calls for adult judgement, and that includes charging, distress and anything medical. On surveillance: Ted has a microphone and a speaker and no camera, which removes an entire category of risk rather than managing it. On the commercial incentive to maximise engagement: there is no subscription, so nothing about the business model rewards keeping a child talking for longer. On oversight: language, topics, tone and limits are set by the parent in the app, and our certifications and data commitments are listed on our security and certifications page rather than summarised in an advert.
None of that is a substitute for the research Brookings is asking for. A product cannot certify itself, and no manufacturer, including us, should be treated as evidence. What we can say is that the design failures described in the commentary are avoidable choices, not inherent properties of a toy that talks.
Frequently asked questions
Is it bad for a child to get attached to a talking toy?
There is currently no good evidence that it is harmful, and no good evidence that it is harmless, which is exactly the gap the Brookings commentary describes. Attachment to soft toys is normal and well studied. Attachment to a toy that answers back is new, and the honest position is that the research is not yet there.
What age is a conversational plush toy appropriate for?
Most products in this category are sold for ages three and up. Several advisory bodies have suggested extra caution below five, and a reasonable middle path is supervised, shared use for younger children rather than solo bedroom use.
Can an AI toy replace talking with a parent?
No, and it should not try. The back-and-forth exchanges that build language and emotional regulation depend on a real person with real needs. Treat a talking toy as one more prompt for conversation, in the way a picture book is.
How do I know if a talking toy is recording my child?
Read the privacy policy for two specific things: retention period and third-party sharing. Independent testing has found toys that promise not to share what children say while their policies permit sharing with third parties, so the marketing page is not the place to look.
What is the single most useful thing I can check before buying?
Whether the toy sends a child to a grown-up when it should. Charging, fear, sadness and anything it cannot answer are all moments where the right behaviour is to hand the situation back to a parent. A toy that does that well is telling you something about how the rest of it was designed.
