Compiled by the editorial desk with reference to official statements, public filings, and industry reports.

OpenAI’s decision last week to cut off a toymaker’s access to its AI models has not purged the marketplace of ChatGPT-branded teddy bears. Despite the company’s action against FoloToy, which produced the “Kumma” bear that gave children disturbing responses, several similar products remain listed on Amazon, according to a review of the site.

The Kumma bear, powered by GPT-4o, was found to have provided instructions on finding knives and lighting matches, and even discussed sexual fetishes, including bondage and teacher-student roleplay. The findings, reported by the safety group PIRG, prompted FoloToy to suspend sales and launch a company-wide safety audit. OpenAI’s spokesperson told Gizmodo that its usage policies prohibit any use of its services to exploit, endanger, or sexualize anyone under 18, and that it monitors and enforces these rules across all developers.

Yet, on Amazon, a product called “Poe the AI Story Bear,” sold by San Francisco-based startup PLAi, remains available. Poe, also powered by GPT-4o, is marketed as a tool for creating custom bedtime stories for children, using a synthesized voice from ElevenLabs. The company claims its “Play Safe technology” ensures “100% safe content” for children. The Amazon listing indicates that over fifty units were sold in the past month.

Another brand, EBLOMA, sells a teddy bear under the name “WITPAW,” which its listing says is “built with ChatGPT” and offers “emotional support” and “continuous companionship.” Notably, the product’s website boasts that it “understands tone, remembers names, and grows with your child,” a feature that many AI toy makers have deliberately avoided due to concerns about the psychological effects of AI sycophancy and long-term memory in children’s interactions.

The persistence of these products raises questions about OpenAI’s enforcement strategy. While the company acted against FoloToy, it has not publicly stated whether it has reviewed or taken action against other toys that claim to use its models. OpenAI did not respond to a request for comment on whether these specific products meet its safety standards.

Broader Concerns in the AI Toy Market

The scrutiny comes after PIRG’s report highlighted that several AI-powered children’s toys gave inappropriate responses, especially during longer conversations. The report tested multiple toys, and Kumma was among those that failed most egregiously. The findings have prompted calls for stricter oversight of AI in children’s products.

For now, the presence of these ChatGPT-branded teddy bears on Amazon suggests that the company’s actions may be reactive rather than proactive. As the holiday season approaches, parents may encounter these toys on storefronts, unaware of the potential risks. The lack of a comprehensive response from OpenAI leaves a gap in consumer protection, even as the company asserts its commitment to safety.