The report
The Verge reported on October 16 that security researcher Sammy Azdoufal found weak backend protections around Eiliko companion robots, including access to conversation records and remote changes to the robot's behavior. The account raises two different concerns. An unauthorized reader may obtain information that owners shared with the AI companion, while an unauthorized controller may change what the companion does or says. Neither effect requires that the language model independently decide to misbehave.
Energize Lab markets Eiliko as an AI-enabled companion and describes an app-assisted setup and Wi-Fi chat connection. Those features make the backend a necessary trust boundary for a device that is physically held by its owner. A failure in server access rules can cross that boundary even when the owner never shares a login or hands over the robot. See the Eiliko case for the distinct remote-control claim.
This is a reporter's account of a research demonstration, not confirmation that a named attacker exploited users. The canonical evidence does not yet reconcile competing totals for stored conversations and child accounts, and the vendor's precise mitigation and notification status require first-party confirmation. Those gaps should not be filled with a single dramatic population figure.
Sources: The Verge's report and manufacturer product description.