When The Verge’s Sean Hollister asked his new Eiliko robot whether their conversation was private, it reassured him. Minutes later, a security researcher on the other side of the world had a transcript of their chat.
It was a controlled demonstration, but a revealing one. The small, wearable AI companion from Energize Lab is designed to talk with its owner. Behind its friendly expressions, researchers found security weaknesses in the services that handle those conversations — including flaws that could also let an outsider change the robot’s AI responses.
The Eiliko security flaws were detailed in The Verge’s 15 October investigation and in vulnerability records published the same day. Nine issues received CVE identifiers. Researcher Sammy Azdoufal believes between 62,000 and 68,000 active robots could have been exposed, although that is his estimate of the affected fleet, not a confirmed tally of hacked devices.
A private conversation retrieved in minutes
Hollister bought an Eiliko, connected it to Wi-Fi and spoke to it. Azdoufal was able to retrieve the conversation transcript remotely during the journalist’s test. The researcher also demonstrated changing how the device replied, including its underlying AI configuration and behavioural instructions.
The demonstration matters because it involved a working robot, rather than an attack described only in technical documentation. A person did not need to enter the home or handle the device to interfere with its connected services. That does not mean criminals have taken over every robot, or that every owner’s conversations were accessed by malicious actors.
The distinction is especially important with a conversational companion. A child might tell it about school, family or worries they would not share with a conventional gadget. Those details can become sensitive information once the conversation leaves the room and reaches a cloud service.
What the nine security vulnerabilities reveal
The published CVE records identify different weaknesses across Energize Lab’s cloud infrastructure. Two are rated close to the maximum on the widely used CVSS severity scale.
CVE-2026-105038 scores 9.8 out of 10. It concerns AI model administration functions that were accessible without authentication, potentially allowing a remote attacker to change the model configuration or switch the model being used. The issue is documented for services associated with Eilik and Eiliko.
The even higher-rated CVE-2026-105037, at 9.9 out of 10, involves a different server component. The record describes a file-access weakness in certain middleware versions through which a registered user with limited privileges could read files, potentially exposing server secrets.
Other issues relate to account and device authentication, AI skills and software updates. The CVEs describe weaknesses with different requirements and potential consequences; their ratings do not establish how often anyone used them for an actual attack.
Why children’s conversations are a particular concern
Azdoufal’s published research figures from a 13 October retest included at least 32,228 children’s profiles and approximately 9.1 million stored conversations. The Verge reported access to roughly 13 million conversations in its own account of the investigation. The two conversation figures should not be treated as a single audited total: they come from different reported snapshots, and Energize Lab has not confirmed the overall numbers.
The records described in the investigation could reveal names, ages, family relationships and personal details disclosed in chats. The Verge also reported concerns about photographs and retained voice recordings. Access to such information was demonstrated by the researcher; that is not the same as proof that each profile has been stolen or used by criminals.
There is another unanswered question. Energize Lab’s privacy policy says voice recordings are encrypted when sent to processing partners and deleted immediately after processing. The reported existence of retained audio calls for a clear explanation of what the service stored, for how long and why.
As with the questions raised by LG smart-TV privacy concerns, what happens to information beyond a connected device can matter as much as its visible controls. The stakes are especially personal when the device is designed to invite a conversation.
Has Energize Lab fixed the problems?
Azdoufal reported that 32 of 40 remotely testable issues were still unresolved when he retested on 13 October. That is a finding from a specific date, not a live assessment of the company’s systems today.
The Verge said it had tried repeatedly to obtain a substantive response from Energize Lab before publishing its investigation. As of 20 October, we have not verified a comprehensive manufacturer update explaining which affected services have been secured, whether exposed information has been removed and what action owners should take.
A routine app update may not be enough to address problems in cloud permissions, account authentication or stored records. Equally, without a current technical retest, it would be wrong to claim that all previously reported flaws are still open. Owners need specific confirmation from the manufacturer.
What Eiliko owners can do now
There is no reason to assume an individual family’s robot has been targeted. Until the remediation status is clear, however, owners can take some sensible precautions:
- Pause connected conversations. Disconnect Eiliko from Wi-Fi or switch it off, particularly if children use it to discuss personal matters. This cannot erase conversations already held in the cloud.
- Avoid disclosing identifying details. Do not share home addresses, school information, passwords or sensitive family circumstances with the robot.
- Secure the connected account. Use a unique password, change any reused password elsewhere and enable two-step verification if Energize Lab provides it.
- Ask what data is held. Contact Energize Lab for an explanation of security repairs and how to access or request deletion of stored conversations and profile information.
- Be alert to unexpected contact. Personal details could make impersonation or fraudulent messages more convincing, although this investigation has not established that such misuse has taken place.
The UK’s National Cyber Security Centre recommends securing smart-device accounts, installing genuine updates and checking manufacturers’ security guidance. A factory reset might clear information from the toy itself, but it should not be assumed to delete records retained on remote servers.
Tech Torque Verdict
The most troubling detail is the gap between what this little robot promises and what the investigation demonstrated. Families are encouraged to talk naturally to Eiliko, yet a researcher was able to retrieve private speech and interfere with the system shaping its replies.
The findings warrant serious concern without overstating them. No published evidence establishes that tens of thousands of robots have been criminally hijacked. What is established is that the exposed services did not provide the protection their users should be able to expect.
Energize Lab should explain what was accessible, what has been fixed and how families can remove information they no longer want stored. Until it does, keeping Eiliko offline is a reasonable precaution.
