The discovery came from an American consumer protection lab, where it became clear that the doll provides dangerous and inappropriate responses to children.
Social media users dubbed it “Chucky GPT,” in reference to the scary doll character, a title that was not an exaggeration according to a “gizmochina” report. The Kumma doll, produced by the Chinese company FoloToy, relies on OpenAI’s GPT-4o model to interact with children, but a PIRG American report revealed disturbing behaviors, including giving dangerous instructions to light matches and handling sexual questions without any oversight.
The group also warned that the permanent microphone could be used for children’s recordings in voice scams. The spread of results on “Reddit” and others accelerated the launch of the “Chucky GPT” nickname, amid speculation about the possibility of hacking the system.
OpenAI responded by suspending the game developer’s account for policy violations, stopping the doll’s access to the company’s services. Subsequently, FoloToy announced the temporary suspension of the doll’s sale and a comprehensive review of its safety, and the doll is displayed on the site as unavailable.
The incident highlights the risks of using powerful language models in games aimed at children without strict controls, turning the doll from an educational friend into a new warning about the dangers of integrating artificial intelligence into games. (Al Arabiya)
