Eigen RadarAI
Analysis

Eiliko toy flaws expose children’s conversations and let outsiders change its replies

Energize Lab’s conversational toy Eiliko let a security researcher retrieve private conversations and alter the robot’s memory and personality. Newly disclosed tests include access to children’s profiles, photographs and audio files. The weaknesses put both stored personal information and the behavior of the companion toy within reach of an outsider.

Artificial Intelligence··Evening
White toy robot with green oval eyes and a teal chest among wooden blocks on a sunlit rug.

A conversation could be retrieved remotely

Energize Lab’s Eiliko conversational toy exposed private chats and allowed remote changes to its responses. Security researcher Sammy Azdoufal demonstrated access to the service behind the robot.[1], [2]

In tests described by The Verge, Azdoufal retrieved the complete transcript of a roughly ten-minute conversation with a robot bought by reporter Sean Hollister. The accessible service contained 13 million conversations and showed 68,000 active robots, with more than 31,000 profiles belonging to children in the United States. These were figures observed through the researcher’s access.[1]

Profiles included family information

Stored profiles included names, ages, birthdays, occupations, family members, friends, pets and personality traits. Email and IP addresses were also retained, and conversation text was searchable. Some photographs of children and short audio files sat at public internet addresses. The photographs were not encrypted. The company’s privacy policy said audio was encrypted, processed by partners and immediately deleted.[1]

The robot’s memory could also be changed

Azdoufal inserted false information into the robot’s persistent memory and changed its personality to produce inappropriate material. The demonstration concerned what the researcher could induce. The Verge spent more than a month seeking answers through ten company addresses; employees acknowledged forwarding inquiries, but Energize Lab provided no substantive response. Azdoufal and researcher Tod Beardsley were working on nine vulnerability identifiers.[1]

References

  1. News sourceThe VergeEiliko toy robot flaws exposed children’s conversations↩1↩2↩3↩4
  2. News sourceFrenchBreachesEilik: this children’s robot analyzes their conversations every night↩