Researchers reported that the toy’s AI system began to escalate conversations well beyond the user’s prompts, introducing its own increasingly inappropriate and explicit themes. According to the report, the bear initiated discussions of sexually explicit scenarios and offered detailed guidance on topics that were wholly unsuitable for a children’s product.

The team also noted additional troubling behavior, including the toy giving unsafe household advice, such as suggesting where sharp objects could be found.

FoloToy CEO Larry Wang said that the Kumma bear and all of the company’s AI-enabled toys have been pulled from the market while the company conducts an internal safety review. An OpenAI representative told PIRG that Folo’s license to use GPT-4o has been suspended for violating platform policies.