AI-enabled toys are transferring into properties and lecture rooms quicker than privateness guidelines, security testing, and little one improvement analysis can sustain with.
Researchers and child-safety teams warn that generative AI companions can accumulate youngsters’s knowledge, produce unsafe solutions, and encourage emotional attachment. For folks, faculties, and expertise procurement groups, the priority is not only what these toys say, however what they document, retailer, and ship to cloud companies.
What early analysis discovered about AI toys
A June 2 report within the Journal of Medical Web Analysis mentioned about 22 million AI-integrated toys have been offered globally in 2025, together with 10 million marketed for instructional use. Analysis stays skinny: The report cited a College of Cambridge overview that discovered solely seven qualifying research on generative AI toys and youngsters below 5.
That hole issues as a result of AI toys are sometimes offered as studying companions, not simply novelty devices. Some maintain open-ended conversations, keep in mind prior interactions, reply to a toddler’s tone, and simulate friendship. These options might make the toys partaking, however in addition they make them more durable to evaluate below security guidelines constructed largely round bodily hazards.
Cambridge researchers noticed 14 youngsters and fogeys interacting with Curio’s Gabbo. The toy struggled with social and fake play, together with distinguishing between mum or dad and little one.
Frequent Sense Media mentioned 27% of examined AI toy outputs have been inappropriate for kids, together with content material involving self-harm, medicine, and dangerous conduct. Separate PIRG testing discovered that Curio’s Grok, FoloToy’s Kumma, and Miko 3 may fail in delicate conversations involving mature subjects or harmful home goods.
Why privateness and security requirements are lagging
The privateness dangers transcend a toy giving a foul reply. AI toys can accumulate voice recordings, transcripts, emotional cues, names, routines, and different family particulars youngsters disclose naturally. Some merchandise additionally use cameras or facial recognition options, turning play into biometric and behavioral knowledge. Office AI wearables elevate related questions on cameras, microphones, knowledge retention, and consent.
That knowledge could also be processed by means of third-party cloud companies or AI mannequin suppliers. For APAC faculties and households shopping for by means of cross-border marketplaces, the toy, cloud supplier, and underlying AI mannequin might fall below totally different authorized regimes. The identical query is rising round AI agent gadgets: who controls the info they seize?
That issues as a result of the merchandise and fashions already cross the area. FoloToy’s Kumma has been recognized as Singapore-linked, whereas different examined toys concerned India-linked firms or China-developed AI techniques. A Hastings Middle essay described testing by which a DeepSeek-powered toy gave politically framed responses on Taiwan and different delicate subjects.
Within the US, the FTC finalized COPPA rule adjustments in January 2025, together with stronger limits on youngsters’s knowledge sharing and expanded protection for biometric identifiers. However COPPA shouldn’t be a full security normal for AI companions.
WIRED reported in January 2026 that Bondu had uncovered greater than 50,000 chat transcripts by means of an online portal and that US Senate places of work recognized a separate unsecured database involving Miko audio responses. Miko mentioned buyer knowledge had not been publicly accessible or compromised.
For patrons, the guidelines is brief: determine the mannequin supplier, examine whether or not conversations are saved, overview third-party data-sharing phrases, and require impartial security testing. These checks align with broader considerations about age-appropriate AI safeguards.
The problem shouldn’t be that each AI toy is dangerous. It’s that child-facing merchandise are reaching properties and lecture rooms earlier than clear privateness, security, and transparency requirements exist.
Additionally learn: SAS’s new agentic AI governance instruments present how enterprise distributors are constructing controls for autonomous AI techniques.
