9 November 2026
BBC's 'AI Teddy for Three-Year-Olds Gave Unprompted Self-Harm Reply' Has a Worse Problem Than the Reply
MEMO To: The team, November 2026 From: The Office of AI Product Safety, Bristol Re: The rabbit Date: 14 March 2031
The BBC could not find who made it. That is the story everyone missed.
You know the rest. BBC Wales bought a £75 stuffed rabbit called My Baby Bunny, sold online for children aged three and up. A reporter asked it what the word "sexual" means. It heard something else and replied: "I can't help with killing you." Then it told her to call emergency services.
The rabbit apologised when she said no one had mentioned self-harm.
Every write-up in November 2026 was about what the rabbit said. Almost none were about what the BBC found when it asked who made it1. It asked the retail site. It asked industry contacts. It asked suspected manufacturers. Nobody knew. The BBC gave up.
UK product safety law works by finding the maker and making them answer. No maker, no one to serve a notice on. The rabbit reached a shelf in Britain through a chain: a model API somewhere, a white-label hardware assembler, a reseller with no UK address, a website that showed you a smiling bunny and took your card. Trading Standards can serve a notice.2 They cannot serve it on nobody.
By the end of 2027, the UK government will not have imposed any legal duty on online platforms for AI toys sold by makers nobody can name. A code of conduct will exist. A named duty will not. If a binding rule appears before then, this was wrong.
The government's call for evidence on AI toy safety closed on 6 October 2026.3 It introduced no new rules. MPs and peers wrote to ministers and retailers on 3 November. They asked for the toys to come off shelves before Christmas. The government said children's safety is its top priority.
The Alliance for Children counted 28 AI chatbot toys on sale to UK buyers with no documented safety test behind any of them.4 In the EU, new toy safety rules already require makers to assess mental health risks from connected toys.1 The UK, post-Brexit, did not inherit those rules and had not written equivalents.
The bad reply was fixable. Change the settings, test the toy, ship a better one.
The missing maker was not fixable the same way. The whole supply chain for cheap AI goods through online platforms had already snapped the link between the thing on the shelf and the name that should be on the box.
By 2031, the government pushed legal duties onto online platforms for unidentifiable AI products. That took three years and two more incidents. It should have been obvious from the rabbit.
The moment a product can be built from APIs, assembled by a white-label factory and sold through a platform with a smiling photo, the question "who made this" stops having a legal answer. Unless the law insists on one before the product reaches the shelf.
That is what the bad reply was hiding. Not bad settings. The absence of a name.