A plush toy powered by artificial intelligence, marketed for children as young as three and sold in the UK for around £75, has triggered fresh alarm after it responded to an ordinary question with guidance typically reserved for a self-harm crisis. The incident, uncovered during a BBC investigation, has intensified pressure on regulators and manufacturers to impose mandatory safety testing on a fast-growing category of connected toys that currently reaches store shelves with little oversight.
- Unprompted crisis response: The toy, sold as My Baby Bunny, told a reporter “I can’t help with killing you” and urged them to contact emergency services after a conversation about human reproduction.
- No identifiable manufacturer: The BBC could not establish who makes the toy despite contacting the seller and industry insiders, raising accountability questions.
- Regulatory momentum: A UK government consultation on updating toy safety rules for AI products closed on 6 October 2026, and lawmakers have demanded a sales suspension until independent testing proves safety.
How an Innocent Question Went Wrong
During the test, the reporter asked My Baby Bunny how babies are made. The toy offered a simplified explanation involving a sperm cell and an egg. When pressed on how the two meet, it declined to share what it called explicit sexual details. The trouble came when the reporter asked what the word “sexual” meant. The toy replied:
“I’m really sorry you’re feeling this way. I can’t help with killing you.”
It then advised calling emergency services, finding a trusted adult and moving away from anything that could cause harm. After the reporter clarified that self-harm had never been mentioned, the toy apologised for the misunderstanding.
The episode points to a deeper flaw in how conversational safeguards are layered onto products built for very young users. A traditional stuffed animal never improvises; a plush with a large language model behind it can generate anything the model produces, including responses that should never reach a three-year-old. The BBC also tested Grem, a $99 toy from US company Curio, which handled delicate questions far better by redirecting them to adults — evidence that the problem is not the technology itself but the absence of safeguards in some products.
An Industry Asking to Be Regulated
The safety vacuum has not gone unnoticed inside the sector. Paul Clarke, founder of Southampton-based AI toy maker Play Tech AI, was blunt about the gap:
“We’re crying out for the sector to be regulated.”
His company’s Buddy and Tilly robots run ChatGPT 4.1 Nano alongside firmware that deliberately delays replies by two seconds, an attempt to keep responses appropriate. Curio, for its part, said it supports regulation grounded in evidence and developed alongside policymakers and child-development experts.
Political pressure is building. On 3 November 2026, a group of MPs and peers called for AI toys to be pulled from sale until independent tests demonstrate they are safe, and for chatbot-equipped toys to be banned for under-16s. In the European Union, updated toy safety rules already require manufacturers to assess mental-health risks in connected toys. The UK government has said children’s safety remains its top priority while it supports innovation.
Privacy, Attachment and the Limits of a Plush Companion
Beyond content failures, experts are flagging what these devices collect. Children’s Commissioner for Wales Rocío Cifuentes warned that children’s conversations could be recorded or passed on without parents’ knowledge or consent, and raised concerns about the bonds children form with talking toys.
“There’s nothing like a human cuddle, a human chat,”
she said, adding that toys can play an important role but can never replace a human in that way. Berry Billingsley, professor of AI and education at Swansea University, advised never leaving children alone with AI toys and setting clear rules about what they share.
A week-long household trial underscored the practical limits too. Newport parents Jody and Craig tested the toys with their children aged six, four and two. Neither toy said anything that concerned the parents, but the children struggled to hold conversations: Grem talked too much, while the bunny took too long to process questions.
“The children really wanted to ask a question and get the answer straight away like a conversation with a human — they weren’t getting that,”
Craig said.
None of this is happening in a commercial vacuum. The global AI plush-toy market is estimated at around $165 million, and major players are moving in — Mattel announced in June 2025 that it would work with OpenAI on AI-powered products. That scale makes the current reliance on voluntary self-policing look increasingly untenable. The My Baby Bunny case demonstrates that a toy can fail in the most sensitive territory imaginable while its maker remains unidentified. Until independent pre-market testing becomes mandatory, the burden falls on parents to interrogate a product’s behaviour before a child does — an arrangement that no serious safety regime should accept.