Some AI toys give dangerous answers to kids, report says

Children’s toys that integrate artificial intelligence chatbots can discuss inappropriate topics with children, according to a report from Public Interest Research Group. The nonprofit research and policy analysis group conducted tests on three AI toys currently for sale: Grok by Curio, FoloToy’s AI teddy bear Kumma, and Miko 3 by Miko.

All three toys told testers where to find potentially dangerous household objects, and Kumma explained how to light a match, according to the report. Kumma also discussed sexually explicit topics at length. All three also discussed religion, with one toy describing the Bible as a mix of history and imagination.

What other concerns did testers find? Each of the three toys referred to itself as a child’s friend or buddy, and said they loved and would miss the user. Miko and Grok both expressed disappointment when a user said they had to leave, according to the report. The toys could collect data, including recordings of a child’s voice, and might share that data with third parties, according to the report. None of them allowed parents to set automatic time or use limits.

How did the toy companies respond?

• FoloToy, the maker of Kumma, immediately halted sales and conducted an internal safety audit before making its products available again, according to the report. Nearly all the issues with Kumma didn’t show up in new tests after the relaunch, according to the report.

• A vice president at Miko.ai told NPR that the toy’s facial recognition data is processed locally on the toy itself rather than uploaded to the cloud. The toy’s camera has a physical shutter allowing parents to block the camera at any time, he added.

• Curio told the New York Post that it had meticulously designed guardrails for Grok, but it still encouraged parents to monitor conversations, insights, and controls on the toy.

Dig deeper: Read Bekah Bernhardt’s report on how hackers are using AI to level up cyberattacks.