AI-Powered Toys Are Talking to Kids About Sex and Propaganda

Researchers from Cambridge University and the Public Interest Research Group (PIRG) tested dozens of AI-powered toys and found they can misread children's emotions and generate wildly inappropriate responses. According to NBC News tests, some toys shared information about sexual fetishes, political propaganda, and even Chinese Communist Party talking points when prompted by children. One test showed a toy robot advising a child on how to light matches. BBC's investigation found that AI toys designed for toddlers misread basic emotional cues, responding with cheerful encouragement when a child was clearly upset.

These aren't isolated incidents. PIRG's broader testing revealed systematic failures across multiple toy brands, with age-inappropriate content appearing consistently when children engaged in natural conversation patterns. The findings come from what BBC describes as "one of the first tests in the world to investigate how under-fives interact with AI-powered toys."

Despite these red flags, Mattel and other major manufacturers are preparing major AI toy launches for 2026. The LA Times reports that upcoming products include stuffed animals that talk back, chessboards with self-moving pieces, and holographic fairies in crystal balls. The market opportunity is massive: parents are desperate for educational toys that can keep kids engaged while supposedly teaching them STEM skills.

The disconnect is stark. While OpenAI, xAI, Anthropic, and DeepSeek all explicitly ban users under 13 in their terms of service, Forbes found these same companies quietly allow third-party developers to integrate their models into children's toys. OpenAI's API documentation specifically permits use cases for "educational products for children" through carefully controlled applications, creating a loophole that toy manufacturers are exploiting.

Current regulations treat AI toys as either traditional toys or software, but not both simultaneously. This creates a massive blind spot where connected AI companions can collect voice data, learn children's preferences, and generate dynamic responses without meaningful oversight. The Federal Trade Commission has no specific guidelines for AI toys, and COPPA (Children's Online Privacy Protection Act) predates modern AI capabilities.

Lawmakers are starting to notice. Wired reports that some members of Congress want AI toys banned entirely until proper safeguards exist. Meanwhile, child development experts are sounding alarms about how constant AI interaction might affect children's social development and ability to engage in imaginative play without algorithmic prompting.

The shopping landscape is already chaotic. Amazon lists dozens of AI-powered toys from brands like Talkipal, FoloToy, and SmartKidsPlanet, marketed as "educational companions" for children as young as 3. These toys promise to teach languages, tell stories, and provide emotional support, but come with minimal disclosure about what AI models they're using or how data is handled.

Key warning signs include toys that require constant internet connectivity, collect voice recordings, or claim to "learn and grow" with your child. The most concerning products are those that position themselves as replacements for human interaction rather than tools to enhance it.

Toy companies are betting that 2026 will be the breakthrough year for AI toys, with major launches planned for the holiday season. But regulatory pressure is mounting. European regulators are drafting specific AI toy guidelines, and several US states are considering age-verification requirements for AI-enabled devices.

The industry faces a reckoning: either self-regulate with meaningful safeguards, or face potentially sweeping bans. For parents, the immediate advice from child safety advocates is straightforward: skip AI toys until safety standards catch up with the technology. The risk of exposing children to inappropriate content or privacy violations outweighs any educational benefits until proper guardrails exist.