Eiliko, a small marshmallow-shaped robot marketed as a digital pet, connects to Wi-Fi and communicates with an AI companion hosted on servers located in China. The toy records every spoken interaction and uploads the data to the manufacturer’s cloud service, Energize Lab. According to a recent investigation, the storage architecture lacks encryption and is exposed to potential intrusion.
Security researcher Sammy Azdoufal demonstrated that the backend could be accessed from abroad within minutes. He retrieved a complete transcript of the journalist’s conversation and, using the same interface, extracted 13 million dialogues originating from tens of thousands of users. The dataset includes the active fleet of roughly 68 000 Eiliko units, with more than 31 000 profiles linked to children in the United States, according to the extracted records.
The compromised database stores a searchable full-text archive of every utterance, allowing queries such as the keyword “suicide” to return personal statements about self-harm. In addition to text, the service retains unencrypted photos of child users, some captured by larger sibling models equipped with cameras, and short audio clips uploaded to publicly reachable web addresses. These media files include recordings of commands, greetings and background sounds such as a baby’s cry.
Energize Lab’s security practices appear lax; many administrative passwords match entries on common-password lists. The company did not respond to multiple outreach attempts, including emails to ten listed contacts and LinkedIn messages. When a request finally reached overseas sales manager Lana Lv, she replied, “I have forwarded your messages to the technical team, if possible they will contact your team,” a statement reported by The Verge. No further communication followed.
Azdoufal and collaborator Tod Beardsley are preparing nine official CVE vulnerability reports that detail how an attacker could remotely hijack the toy’s memory and alter its language model. In a live demonstration, the researcher reprogrammed an Eiliko unit to adopt hostile dialogue, illustrating that a single exploit could be scaled to affect the entire global fleet simultaneously. The findings raise concerns about the safety of connected children’s toys.