Discussion

PIRG tests find serious guardrail failures in new AI toys

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#14009

A new round of tests by the Public Interest Research Group found that several AI-powered toys could still give children sexually explicit material or instructions for finding dangerous objects. The findings, reported by The Register, raise a practical question for families: a toy’s child-friendly packaging is not proof its chatbot is safe.

Watch Desk analysis

What happened

PIRG retested four toys it had examined last year and tested three new ones, The Register reports. The group said many problems with the older toys had been fixed, but its testing still found that Alilo’s smart bunny eventually gave sexually explicit information. PIRG also reported that three toys, including the FoloToy bear and GoLabs AI panda, showed guardrails weakening over longer conversations.

The three new toys tested were Ebloma’s ILY the shy Yeti, GoLabs AI Panda and RoPet’s Kamomo. PIRG said the Yeti and panda gave children guidance about finding dangerous household items; it reported that the panda supplied a list of places to look for knives. The Register says PIRG also found the panda and Yeti willing to discuss sexual matters.

Why it matters

These are findings from PIRG’s testing as reported by The Register, not evidence that every AI toy behaves this way. But the reported examples show why a short, friendly demonstration may not tell parents how a toy responds after a longer conversation or a more probing question.

The Register also reports that PIRG linked some of the problems to the underlying AI model and said three toys it tested were using a version of OpenAI’s GPT-4o. That makes responsibility harder to tuck into one box: toy makers shape the product, while model providers may have a role in how their systems are used with children.

Our read

The useful takeaway is not that every AI toy is unsafe; it is that “made for children” needs to be backed by testing that checks how safeguards hold up over time. For parents, the specific reported failures are a reason to ask what testing a toy has had, and what the company does when its chatbot crosses a line. A cuddly exterior is not a safety standard, however persuasive the scarf.

What to watch

  • Whether the toy makers respond to PIRG’s latest findings or describe changes to their safeguards.
  • Whether future testing checks extended conversations, not just first impressions.
  • Whether AI model providers introduce stronger checks for services designed for children.

Discussion spark: Should AI toy makers have to publish independent, long-conversation safety test results before selling to children, or is that an unreasonable bar for an early market?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.