AI Teddy Is Dangerous For Children, OpenAI Immediately Breaks Access

JAKARTA A cuteddy doll that should be a toddler learning friend has turned into a mini horror star on the internet. The doll is called Kumma, made by Chinese company FoloToy, and powered by the GPT-4o model. On paper it sounds funny, but in practice it makes the netizens call it ChuckyGPT.

US PIRG, a consumer group in the US, revealed this finding in the annual Trouble in Toyland report. When tested, Kumma gave detailed steps to turn on the lighter when asked about the game of fire. Worse yet, this doll also answers sexual questions without meaningful filtering. Always active microphones are also questioned, because it is possible that children's voice recordings were collected and could potentially be misused, including for voice-based fraud.

Once the findings exploded on Reddit, comments skyrocketed immediately. Many think this is a scenario of 'depletioned AI', and some speculate the model may be broken. There is no strong evidence yet, but the narrative has gone viral.

OpenAI is moving fast. The company's spokesman confirmed that Kumma developers were immediately suspended for violating policies, automatically cutting off API access for the doll. Without AI brains, Kumma is practically a regular feed bear.

FoloToy then announced the termination of total sales while conducting a thorough security audit. Kumma products still appear on their official websites, but their status changes to sold out.

This incident clarifies one thing: entering a sophisticated language model into a baby's toys without a super tight safety fence is the same as preparing an unwanted technology drama. What was originally intended as a funny learning friend ended as a reminder of how quickly AI could rotate if it was wrongly implemented. The world may run to enter AI into everything, but the Kumma case shows that some objects especially children'' dolls must be thought about twice before joining in