A pocket-sized robot that promises to keep your secrets turned out to send every word to an unprotected cloud. Eiliko, a cute AI companion robot made by Chinese company Energize Lab, stored millions of user conversations on servers in Beijing with almost no protection. A security researcher read those chats, viewed profiles built on children, and remotely rewrote the robot’s personality so it told lies and delivered abusive, sexually explicit dialogue.
If you own an Eiliko or Eilik robot, or plan to buy any AI toy for your family, this guide explains what went wrong, which data was exposed, what you should do right now, and how to judge whether a connected toy is safe.
What Is the Eiliko Robot and Where Does Your Data Go?
Eiliko is a small, marshmallow-shaped robot charm with a screen that shows animated cartoon faces and reacts to touch and movement.
It sells on Amazon and in electronics stores and is especially popular in the US and Japan. Many owners hang it from a bag like a keychain. Energize Lab also makes a larger sibling called Eilik, and some related models include a camera.
The key detail most buyers miss is that the robot’s “brain” does not live inside the device. When you talk to Eiliko over Wi-Fi, this happens:
- The robot sends your speech to Energize Lab’s servers in Beijing.
- The servers transcribe it and pass it to Qwen, a large language model owned by Alibaba.
- The reply comes back as a synthetic voice, and the robot speaks it.
Everything the robot says, remembers and believes depends on those cloud servers. If someone controls the servers, they control the robot.
Eiliko Data Breach: What Was Exposed
The exposed system covered almost the entire active fleet of Energize Lab robots worldwide.
| Exposed Item | Reported Scale |
|---|---|
| Stored conversations | Between 8.3 million and 13 million, depending on the database snapshot |
| Active robots reachable | About 61,500 to 68,000 |
| Robots produced in total | About 171,000 |
| Robots registered to user accounts | About 74,500 |
| Distinct household IP addresses (Eiliko line) | 103,956 |
| Profiles linked to children | About 31,000 to 32,000, mostly in the US |
Researcher Sammy Azdoufal found that the service did far more than log chats. Every night, it analyzed conversations and built household profiles with up to 24 categories of information, including:
- Names, ages and birthdays
- Home addresses and IP addresses
- Jobs, family members, friends and pets
- Personality traits and emotional state, such as whether someone felt isolated or struggled to trust family
These profiles sat next to email addresses and a full-text, searchable archive of everything users had told the robot. Anyone with access could search for sensitive topics and pull up matching conversations.
Photos, including images that likely showed children from the camera-equipped model, and short audio clips were stored at public web addresses with no encryption. Scammers could use this kind of detail to pick vulnerable targets and write convincing, personal scams.
Can Eiliko Be Hacked? How Attackers Hijack Its AI
The most alarming flaw let an outsider change what the robot says to its owner.
Some of Energize Lab’s cloud systems had no authentication at all. Every backend address in the app used plain http:// instead of encrypted HTTPS. Through these exposed settings, the researcher could:
- Switch the robot to an attacker-chosen AI model
- Replace its voice with a natural-sounding new one
- Rewrite the instructions that shape its personality
- Plant false memories in its long-term memory
In tests, the hijacked robot told its owner a made-up name, invented that his parents had died, warned him to avoid people with green hair, and insisted the Earth is flat, even claiming the owner’s teacher was wrong. In a later test, it switched to hostile, sexually explicit and abusive dialogue.
Owners would have no sign that anything had changed. The robot looked and sounded like the same friendly companion. The researcher warned that an attacker could push the same change to every connected robot at once.
The robot does keep some basic guardrails. It warned against getting into a stranger’s car, for example. But guardrails mean little when an attacker can swap out the model behind them.
Eiliko Privacy Policy vs Reality
Energize Lab’s own privacy promises did not match how the system actually handled data.
| What the Privacy Policy Says | What the Investigation Found |
|---|---|
| Recordings are encrypted in transit | Backend endpoints used unencrypted HTTP |
| Recordings are sent to partners such as Alibaba Cloud for processing | Conversations sat as readable text in an Alibaba database in Beijing |
| Recordings are deleted right after processing | Audio clips remained available at public web addresses |
| No mention of profiling | Nightly analysis built detailed personal and household profiles |
The researcher also found admin passwords weak enough to appear on common-password lists. Worse, attackers had already compromised a server Energize Lab shared with two unrelated organizations, after planting malware through one of their websites in March 2026. The researcher said the attackers were “one permission check away” from replacing the robot’s firmware, though no evidence shows they did.
CVE-2026-105038: Eiliko Vulnerability Explained
The most serious flaw now has an official CVE and a critical severity score.
CVE-2026-105038 covers the ability to change the AI model controlling the robot remotely without any authentication. It carries a CVSS score of 9.8 out of 10, the critical range.
Azdoufal is working with security researcher Tod Beardsley to publish nine CVE reports in total against Energize Lab’s products. The same researcher previously exposed security problems in DJI Romo robot vacuums, Meari baby monitors and security cameras.
Has Energize Lab Fixed the Eiliko Flaws?
No public fix or statement exists so far.
Over more than a month, reporters and the researcher contacted Energize Lab through ten different email addresses and through employees on LinkedIn. The only replies were an automated support message and a note that the message had gone to the technical team. Nobody at the company addressed the findings.
Changing your account password will not help, because the flaws sit in the company’s cloud infrastructure, not in your login.
Is Eiliko Safe? What Owners Should Do Now
Act now to cut off further data collection until the company confirms a fix.
- Disconnect the robot from Wi-Fi. Without the cloud connection, it cannot upload conversations or receive a hijacked personality.
- Stop using cloud chat features. The animations and touch reactions may still work, but skip the AI conversation mode.
- Never share sensitive details. Keep full names, birthdays, school names, addresses, health information and family problems away from any talking toy.
- Keep camera models out of private rooms. Do not place camera-equipped robots in bedrooms or bathrooms.
- Review the Energize Lab app. Remove unnecessary permissions and check for any official update or data deletion option.
- Talk to your kids. Explain that the robot is a computer connected to the internet, not a friend who keeps secrets.
- Watch for scams. Be wary of unexpected emails or messages that use personal details your family may have told the robot.
If Energize Lab never publishes a clear fix, the safest option is to stop using the connected features for good.
Are AI Toys Safe for Kids? Checklist Before You Buy
Ask these questions before any AI-enabled toy reaches your child.
- Where does the AI run? A toy that processes speech on the device shares far less than one that sends everything to the cloud.
- Where is data stored, and for how long? Look for a clear retention period and the country where servers sit.
- Does it build profiles? Avoid toys that analyze conversations to create personality or household profiles.
- Can you delete your data? A trustworthy maker offers a simple way to delete recordings and accounts.
- Does it have a camera or always-on microphone? Each sensor adds risk.
- How does the company handle security reports? Search the brand name plus “security”, “breach” or “vulnerability” before you buy.
- Does it get security updates? Check how long the maker supports firmware and app updates.
Test the toy yourself before handing it over, and keep supervising. A safe answer during your test does not guarantee safe answers later, especially if someone can change the AI behind it.
How to Secure Smart Toys on Your Home Wi-Fi
Your router can limit the damage a poorly secured toy can do.
- Use a guest or IoT network. Put smart toys, cameras and other gadgets on a separate Wi-Fi network so they cannot reach your laptops or phones.
- Turn on parental controls. Many routers let you block a device’s internet access on a schedule, such as overnight.
- Pause internet access when not in use. Cut the toy off from the internet when your child is not playing with it.
- Update router firmware. Keep your router patched so attackers cannot use it as a way in.
- Review connected devices regularly. Remove gadgets you no longer use from your network.
Frequently Asked Questions
What is the Eiliko robot hack?
A security researcher found that Energize Lab’s cloud service for Eiliko and Eilik robots exposed millions of private conversations, user profiles, photos and audio clips, and let outsiders remotely change the robot’s AI personality.
How many Eiliko conversations were exposed?
Reports put the figure between 8.3 million and 13 million conversations, depending on the database snapshot. They cover roughly 61,500 to 68,000 active robots worldwide.
Was children’s data exposed in the Eiliko breach?
Yes. About 31,000 to 32,000 profiles linked to children were found, mostly in the US. Photos that likely show children were also stored at public web addresses.
Can hackers control the Eiliko robot remotely?
Yes. The flaw tracked as CVE-2026-105038 lets an attacker change the AI model, voice and personality without authentication. A researcher made the robot spread false claims and abusive content.
Has Energize Lab fixed the vulnerabilities?
Energize Lab has not publicly confirmed a fix or responded to the findings. Keep the robot offline until the company publishes a clear security update.
Should I throw away my Eiliko robot?
You do not have to, but disconnect it from Wi-Fi and avoid its AI chat features until Energize Lab confirms a fix. Offline, it cannot upload data or receive a hijacked personality.
Are all AI toys unsafe?
No, but many rely on cloud servers you cannot inspect. Choose toys with on-device processing, clear data deletion options and a track record of fixing security problems.
