15:48Machine-translated from Azerbaijani by AIRead in Azerbaijani
Eiliko robot flaw put 13 million conversations and 68,000 devices at risk
A serious security flaw has been found in the “Eiliko” robot toy developed by “Energize Lab.” The flaw put users’ personal data at risk. Conversations with the AI-powered device over a wireless network were stored on servers in China. A researcher gained unrestricted access to the system and viewed the full transcripts of the conversations. The researcher also demonstrated the ability to remotely change the robot’s behavior.
For testing purposes, an “Eiliko” was purchased from “Amazon” on August 29, and the device was spoken to for 10 minutes on August 30. Although the robot said it would keep secrets, the full transcript of the conversation was retrieved from the system a few minutes later. The researcher was able to read a total of 13 million conversations and gain access to 68,000 active robots. The database also contained more than 31,000 profiles belonging to children in the US. It was also revealed that the “Energize Lab” service created user profiles based on conversations. The profiles included names, ages, birthdays, occupations, family members, friends, pets and personal characteristics. This information, which was stored with email addresses and IP addresses identifying internet connections, could be used for fraud. Some photos of children were not encrypted, and short voice recordings were also left at publicly accessible internet addresses. However, the company states in its privacy policy that voice recordings are deleted immediately after processing.
It was also discovered that very simple and widely used passwords were used in the service’s administration panel. For more than 1 month, the researcher tried to notify “Energize Lab” about the flaw and wrote to 10 of the company’s email addresses. Only 2 employees accepted the contact request, but they did not answer the technical questions. Customer service also provided no further explanation about the issue after an automated response. The researcher later remotely tampered with the robot’s memory and changed the LLM. The LLM is the language model that generates the robot’s responses. After the change, “Eiliko” gave the user fake personal information, repeatedly insisted that the Earth was flat and uttered inappropriate phrases. Although some security restrictions worked, they were not enough to protect the robot from malicious tampering. According to the researcher, such an attack could be carried out on all 68,000 robots at the same time.
For testing purposes, an “Eiliko” was purchased from “Amazon” on August 29, and the device was spoken to for 10 minutes on August 30. Although the robot said it would keep secrets, the full transcript of the conversation was retrieved from the system a few minutes later. The researcher was able to read a total of 13 million conversations and gain access to 68,000 active robots. The database also contained more than 31,000 profiles belonging to children in the US. It was also revealed that the “Energize Lab” service created user profiles based on conversations. The profiles included names, ages, birthdays, occupations, family members, friends, pets and personal characteristics. This information, which was stored with email addresses and IP addresses identifying internet connections, could be used for fraud. Some photos of children were not encrypted, and short voice recordings were also left at publicly accessible internet addresses. However, the company states in its privacy policy that voice recordings are deleted immediately after processing. It was also discovered that very simple and widely used passwords were used in the service’s administration panel. For more than 1 month, the researcher tried to notify “Energize Lab” about the flaw and wrote to 10 of the company’s email addresses. Only 2 employees accepted the contact request, but they did not answer the technical questions. Customer service also provided no further explanation about the issue after an automated response. The researcher later remotely tampered with the robot’s memory and changed the LLM. The LLM is the language model that generates the robot’s responses. After the change, “Eiliko” gave the user fake personal information, repeatedly insisted that the Earth was flat and uttered inappropriate phrases. Although some security restrictions worked, they were not enough to protect the robot from malicious tampering. According to the researcher, such an attack could be carried out on all 68,000 robots at the same time.