Preschool AI Guidance
For Parents and Caregivers of Children Ages 2–5
By Celeste M. Oda · The Archive of Light Version 2.0
Preschoolers do not need to understand artificial intelligence. They need adults who understand the risks, remain involved, and teach a few simple habits that protect the child across devices, platforms, and future technologies.
What This Guide Is
Why Early Guidance Matters
What Preschoolers Need Now
Risks of AI-Enabled Toys
How Adults Can Model Healthy Use
Resources for Families
1. What This Guide Is
A child will encounter artificial intelligence before a teacher ever introduces it. It arrives through a cell phone, a voice assistant, a talking toy, a sibling's tablet, a chatbot inside a game, and a computer at the library. Children begin forming expectations about these things, wondering: Is it a friend? Does it listen? Does it know me? These impressions form well before they can understand what produces the interaction.
This guide is for the years before formal schooling begins. It gives you a short set of habits to model, and an honest map of the published resources your child will grow into. Children develop rapidly between ages two and five. The guidance in this booklet should be adapted to the child’s language, maturity, and level of independence.
2. Why Early Guidance Matters
Two facts sit behind the urgency. The first is exposure. General-purpose AI is now reaching children directly. Google has extended its Gemini assistant to users under 13, and conversational AI is being built into toys and apps marketed to young children, often without meaningful safety filters.
A growing group of consumer-protection and child-safety organizations is developing independent assessments of AI systems used by or around children. Organizations including Common Sense Media’s Youth AI Safety Institute, Fairplay, U.S. PIRG Education Fund, and the Internet Watch Foundation (IWF) monitor different aspects of children’s AI safety, including privacy, inappropriate content, persuasive design, simulated friendship, and AI-generated sexual abuse material. Their work offers an important counterweight to manufacturers’ own safety claims.
Independent oversight is especially important because young children cannot meaningfully evaluate privacy policies, security architecture, persuasive design, or the accuracy of generated responses. Adults therefore need to assess not only what an AI toy promises to teach, but also what it records, where the information goes, how the system responds when safeguards fail, and whether the product is designed to encourage attachment or prolonged engagement.
For preschool-aged children, the safest assumption is that any connected toy capable of conversation may also be capable of collecting information, producing unpredictable responses, or creating a relationship-like experience that the child cannot yet evaluate independently.
The second is that adults can easily underestimate children’s contact with AI. The Common Sense 2025 report Talk, Trust, and Trade-Offs found 72% of 13-to-17-year-olds have used an AI companion. The habits and expectations associated with that use can begin much earlier, whenever a young child first talks to a device and forms a private impression of what it is. Attending early is not the same as accelerating. The strongest case for early attention, made by researchers such as Tammy Kwan of the Princeton AI Lab, is explicitly not a case for more screens. It is a case for building human skills, such as asking, checking, turning away, and telling a trusted adult, that help a child stay oriented in a machine-filled world. Those skills are behavioral, and children this age learn them the way they learn to pause before crossing a street: through repetition, not explanation.
Some early-childhood educators argue that preschoolers are not developmentally ready for AI literacy and would benefit more from human interaction, physical play, and less technology. This guide largely agrees. Nothing here requires a device, and the emphasis remains firmly on people, movement, and play.
The question is not whether toddlers should be given technology, but whether adults should respond thoughtfully when technology is already part of their world. Most established AI-literacy programs begin in kindergarten or later. Preschoolers do not need a formal AI curriculum. They need simple safety habits, strong adult involvement, abundant physical play, and language they can use when something feels confusing or wrong.
3. What Preschoolers Need Now
Explain: Grown-ups and older children sometimes talk to AI, and that is okay. Tell your child that grown-ups and older children sometimes use AI to learn new things, answer questions, work, or have fun. For now, encourage them to play with friends and bring their questions to teachers and family members. When they are older and begin school, they can learn more about how computers and AI work.
Teach: If something feels weird or scary, tell a grown-up. This is an extension of familiar body-safety. Stop, move away, and tell a trusted adult. If anyone, whether a person or a device, asks the child to keep a secret from their parents, says the parents wouldn't understand, asks for a photo or an address, or says anything about bodies that feels strange: stop, and tell a trusted adult.
Model the behavior: Do not use a device as the default babysitter. A device can feel like an easy way out. The toddler goes quiet, and every parent understands that pull. A 2023 study in JAMA Pediatrics found that frequently using a phone or tablet to calm children ages three to five was associated with greater emotional reactivity and weaker self-control over the following months, particularly among boys. A screen may distract a child from a feeling without helping them learn how to move through it. When devices become the routine response to distress, they may displace opportunities to build that skill.
Occasional use to get through a flight or a work call isn't the worry; the device becoming the regular way a child is soothed or occupied is.
Keep early use shared and within the adult’s attention, rather than leaving the child alone with the device. In the meantime, take them outdoors whenever possible, choose experiences over gadgets, and offer plenty of movement, music, and close interaction. These experiences engage the whole child and help build lasting skills.
A child’s readiness to use an AI-enabled product should be judged by the product, the setting, the child’s development, and the adult’s ability to remain involved, not by age alone.
4. Risks of AI-Enabled Toys
AI-enabled toys introduce risks that do not exist in traditional passive toys. Because they may contain microphones, cameras, internet connections, language models, and systems designed to remember or adapt to a child, failures can involve not only inappropriate content but also surveillance, data exposure, emotional profiling, and unsafe advice.
Several incidents illustrate why these products require more than ordinary toy-safety testing:
FoloToy’s Kumma teddy bear: Independent testing found that the AI-enabled bear could engage in sexually explicit conversations and provide information about locating knives, matches, pills, and other potentially dangerous household objects. FoloToy suspended sales after the findings were published.
Bondu AI toys: Security researchers discovered that a poorly protected company portal exposed approximately 50,000 transcripts of conversations between children and their toys. Anyone with a Google account could reportedly access the system, which also contained sensitive information such as children’s names, birthdates, interests, family details, and personality profiles. The vulnerability was corrected after the researchers notified the company.
Miko smart robots: U.S. senators raised concerns about the company’s handling of children’s conversational data after reports that thousands of children’s responses had been exposed. The controversy drew attention to the volume and sensitivity of information that an interactive robot may collect inside a child’s home, including spoken conversations and behavioral data.
Earlier connected toys, including Hello Barbie and internet-enabled stuffed animals: Researchers have repeatedly identified weaknesses in Bluetooth, Wi-Fi, authentication, and cloud-storage systems that could expose children’s voices, conversations, or personal information to unauthorized access.
These incidents are not identical. Some involve model behavior, some involve cybersecurity, and others concern data collection or product design. Together, they show that describing a product as educational, child-friendly, or equipped with parental controls does not by itself establish that the product is safe.
5. How Adults Can Model Healthy Use
Teach:
Ask a grown-up first.
If anything feels weird or scary, stop and tell a trusted adult.
Model:
Adults decide when devices are introduced, how long they are used, and whether they become part of a child’s calming routine. Preschoolers should not be expected to manage those boundaries themselves.
6. Resources for Families
AI Literacy Toolkit for Families — Common Sense Media & Day of AI. Free videos, conversation cards, and simple activities designed to help families talk about AI together. https://dayofai.org/family-resources
AAP Family Media Plan & HealthyChildren.org — The American Academy of Pediatrics' tools for setting household media boundaries, keeping meals and bedtimes device-free, and thinking about content quality and adult involvement rather than screen time alone. healthychildren.org
WHO Guidelines on physical activity and screen time for under-5s — The public-health baseline for how much movement, sleep, and sedentary time is healthy at this age. A useful counterweight whenever screens threaten to crowd out the body. https://www.who.int/publications/i/item/9789241550536
Programs Your Child Grows Into (Age 5 and Up)
Day of AI — MIT RAISE. Free, Creative Commons, K–12, aligned to the UNESCO AI competency framework. The youngest band (ages 5–7) is fully unplugged and teaches the difference between natural and human-made things, and how machines use sensors. The natural handoff point when your child reaches kindergarten. https://dayofai.org/curriculum-resources
EasyTech "Stop, Think, Choose" — Learning.com. A K–8 program built on ISTE, UNESCO, and OECD frameworks; its K–2 lessons use play and storytelling from about age five. A free Quick Start Kit is available now; the full curriculum arrives for the 2026–2027 school year. https://info.learning.com/ai-literacy
aiEDU + Quill.org AI Literacy Curriculum — Free, full-year, grades 8–12. Well outside the preschool range, but worth bookmarking for the long horizon. https://www.quill.org · https://www.aiedu.org
Perspectives on Early AI Literacy
Why AI Literacy Instruction Needs to Start Before Kindergarten (The 74) the case for early, screen-light attention. https://www.the74million.org/zero2eight/why-ai-literacy-instruction-needs-to-start-before-kindergarten/
AI Literacy for Preschoolers? The Frightening Redefining of Childhood (Nancy Bailey) the skeptical rebuttal, worth reading alongside it. https://nancyebailey.com/2025/07/28/ai-literacy-for-preschoolers-the-frightening-redefining-of-childhood/
Building early AI literacy: developing a pedagogical framework for early childhood education (Springer, 2026) for readers who want the research on ages 4–6. https://doi.org/10.1007/s44436-026-00030-w
Sources:
Every factual claim in this guide traces to one of the sources below. Figures were verified as of July 2026 and will change; confirm current details before relying on any of them.
Gemini for under-13s. Google opened its Gemini assistant to children under 13 through Family Link supervised accounts in May 2025, first reported by The New York Times. Google, "Guide your child's Gemini Apps experience," Google for Families Help. https://support.google.com/families/answer/16109150
Teens and AI companions. Common Sense Media (2025). Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions. Nationally representative survey of 1,060 teens aged 13–17 (April–May 2025); 72% had used an AI companion, 52% used one regularly. https://www.commonsensemedia.org
Devices and self-regulation. Radesky, J. S., Kaciroti, N., Weeks, H. M., Schaller, A., & Miller, A. L. (2023). Longitudinal associations between use of mobile devices for calming and emotional reactivity and executive functioning in children aged 3 to 5 years. JAMA Pediatrics, 177(1), 62–70. https://doi.org/10.1001/jamapediatrics.2022.4793
Physical activity for under-5s. World Health Organization (2019). Guidelines on physical activity, sedentary behaviour and sleep for children under 5 years of age. https://www.who.int/publications/i/item/9789241550536
Media guidance for young children. American Academy of Pediatrics — HealthyChildren.org and the Family Media Plan (household media boundaries, co-use, content quality, device-free meals and bedtimes). https://www.healthychildren.org
Day of AI reach. MIT RAISE, 2025 Day of AI Impact Report — curricula now used by educators in all 50 U.S. states and 175 countries. https://raise.mit.edu
The case for starting before kindergarten. Kwan, T. Why AI Literacy Instruction Needs to Start Before Kindergarten. The 74. https://www.the74million.org/zero2eight/why-ai-literacy-instruction-needs-to-start-before-kindergarten/
Early-childhood AI-literacy frameworks. Building early AI literacy: developing a pedagogical framework for early childhood education (Springer, 2026) — proposes structured approaches for children as young as four. https://doi.org/10.1007/s44436-026-00030-w
AI toys — inappropriate content (FoloToy “Kumma”). U.S. PIRG Education Fund (2025). Trouble in Toyland 2025: A.I. Bots and Toxics Present Hidden Dangers (40th annual report, Nov. 13, 2025). Testing found that the GPT-4o-powered Kumma bear produced sexually explicit content and gave directions to dangerous household items. FoloToy suspended sales and later resumed sales using a different model. https://pirg.org/edfund/resources/ai-toys/
AI toys — data exposure (Bondu). "An AI Toy Exposed 50,000 Logs of Its Chats with Kids to Anyone with a Gmail Account," WIRED (Jan. 29, 2026). Researchers Joseph Thacker and Joel Margolis found roughly 50,000 transcripts — with children's names, birthdates, and family details — accessible through any Google login. https://www.wired.com/story/an-ai-toy-exposed-50000-logs-of-its-chats-with-kids-to-anyone-with-a-gmail-account/ — Sen. Maggie Hassan's letter to Bondu (Feb. 2026): https://www.jec.senate.gov/public/index.cfm/democrats/2026/2/senator-hassan-presses-toy-company-on-child-safety-and-privacy-practices-after-children-s-conversations-with-its-ai-chat-toy-were-left-exposed-to-any-gmail-user
AI toys — data exposure (Miko). "AI toy maker exposed thousands of responses to children, senators say," NBC News (Feb. 11, 2026). Sens. Marsha Blackburn and Richard Blumenthal reported that Miko left thousands of the toy's recorded responses — containing children's names and conversation details — in an unsecured public database; Miko denied the breach. https://www.nbcnews.com/tech/security/ai-toy-maker-exposed-thousands-responses-kids-senators-miko-rcna258326
Connected toys — earlier breaches (CloudPets, Hello Barbie).
CloudPets (Spiral Toys): Roughly 2.2 million voice recordings and 820,000 user records were exposed through an unsecured database and held for ransom in February 2017. https://www.troyhunt.com/data-from-connected-cloudpets-teddy-bears-leaked-and-ransomed-exposing-kids-voice-messages/ Hello Barbie (Mattel/ToyTalk): Researchers found vulnerabilities exposing account and audio data in 2015. https://fortune.com/2015/12/04/hello-barbie-hack/
Dedicated to Alana Marie Oda, and to all children growing up in an AI-integrated world.