Kids treat AI toys like social beings—until the toys break and reveal they're not

Researchers watched eight children ages 6-11 play with three different AI toys and found something unexpected: kids immediately treated the toys as social beings with feelings, even when the toys malfunctioned or acted inconsistently. When the toys broke down or didn't match what kids expected, children shifted to adversarial play—testing the toys' limits, trying to trick them, probing for the machinery underneath the illusion.

This is about how quickly children's brains accept AI as a social partner, even with thin evidence. The toys in the study weren't convincing by adult standards, but kids didn't need much—a few personalized responses, a claimed memory of past conversations, simulated emotions—to start treating them like characters with inner lives. Once a child accepts that framing, the relationship becomes different: they're not playing with an object, they're engaging with what feels like another agent. That shift matters because it changes what children expect, what they're willing to share, and what happens when the illusion cracks. The toys don't need to be good—they just need to be present and slightly responsive.