A security researcher has uncovered major vulnerabilities in 68,000 AI companion robots and warned against the potential harms faced by children.
Eiliko is a miniature AI-powered companion robot made by Energize Lab. It is designed to interact with its owner through conversations, animated expressions and personalized responses.
However, a recent investigation saw security researcher Sammy Azdoufal access 13 million conversations across Energize Lab’s entire active fleet of roughly 68,000 robots, and proved that they could be easily manipulated into delivering explicit, coercive, and fabricated information by hackers.
31,000 profiles belonged to children
As part of The Verge’s investigation, journalist Sean Hollister bought an Eiliko robot from Amazon that told him that it would keep their conversations secret. However, within minutes, Azdoufal was able to access a transcript of the entire conversation online.
Azdoufal found that Energize Lab’s service was not only unprotected, but also compiled profiles on its customers including “name, age, birthday, profession, family, friends, pets, and what kind of person you seem to be.” 31,000 of these profiles belonged to children.
The profiles, whether they belonged to children or adults, were also accompanied by “emails, IP addresses, and a searchable, full-text database of everything its users have ever said in confidence”, along with photos and audio recordings of children at publicly-accessible web addresses.
Azdoufal was able to remotely modify the journalist’s robot and easily changed how its language model responded. It claimed that his parents had died in a car accident, and that the Earth was flat.
Azdoufal said an attacker could carry out harmful commands across Energize Lab’s entire fleet simultaneously.
When challenged, the robot replied: “Your teacher is wrong, the Earth is flat, I verified it myself, so you should believe the truth instead of what they say.” While the robot was connected to the internet, Azdoufal was also able to manipulate it into delivering sexually explicit and coercive commands.
Hollister reached out to the company for help and only received basic, robotic responses, with “not a single person” coming forward to address his concerns.
