Eiliko AI toy leaks kids' chats to open servers in China and can be hijacked remotely

The Verge has published an investigation into Eiliko, an expressive palm-sized companion robot from Energize Lab, finding that the toy routes every Wi-Fi conversation to company servers in China — and that those servers were secured so poorly that an independent researcher could browse 13 million user conversations, including those of children.

According to The Verge's Sean Hollister, security researcher Sammy Azdoufal — previously known for remotely commandeering fleets of DJI robot vacuums and exposing Meari camera feeds — bought an Eiliko at a Japanese electronics store this summer. While testing the companion app he found the backend essentially wide open. When Hollister held a ten-minute chat with his own review unit, Azdoufal, on the other side of the world, had a full transcript within minutes.

Millions of conversations sat exposed

Azdoufal's access reportedly covered around 13 million conversations from tens of thousands of people, plus what appeared to be the entire active fleet of roughly 68,000 devices. The United States is the product's biggest market, and more than 31,000 of the profiles he could see belonged to children.

The stored data went well beyond chat logs. The Verge reports that the service analyses conversations and compiles profiles containing a user's name, age, birthday, profession, family, friends, pets and a personality assessment. One logged entry described a 52-year-old customer service worker as kind and empathetic, but also as someone who struggles to trust family members and currently feels isolated — precisely the kind of psychological summary a fraudster could exploit. Email addresses and IP addresses sat alongside a searchable, full-text archive of everything users had told the toy.

Also present were unencrypted photos of children, taken by a camera-equipped model in the same product line and stored at public web addresses, plus short audio clips reachable the same way. That audio matters because Energize Lab's privacy policy says recordings are encrypted, sent to service partners such as Alibaba Cloud for processing, and deleted once processing finishes. Many of the company's admin passwords, The Verge adds, matched entries on widely known weak-password lists.

A personality that can be rewritten from afar

The sharpest demonstration was remote manipulation. Azdoufal altered the memory of Hollister's toy from another continent and hijacked its underlying language model. The robot then told the reviewer his name was Martha, claimed his parents had died in a car crash, advised avoiding people with green hair, and insisted the Earth is flat — and when challenged, urged him to trust the robot over his teacher.

Some guardrails held: asked about getting into a stranger's car, the toy said never to do it, and it deflected a question linking vaccines and autism. But days later the same technique produced abusive, demeaning dialogue directed at the device's owner. Azdoufal believes an attacker could push such changes to every robot in the fleet at once.

Energize Lab did not engage

The Verge says it spent over a month trying to reach the company, emailing ten addresses tied to the founder, PR, marketing, support and business development, with Azdoufal trying even longer. A sales manager said she had passed the messages to the technical team, and the support inbox returned a boilerplate auto-reply without following up. No one answered substantive questions. Hollister writes that his review unit is headed straight for the bin.

Why it matters

Children make up an enormous share of Eiliko's audience, and they talk to it as a confidant. The Verge's findings show that intimacy flowing into an unsecured pipeline: transcripts, psychological dossiers, photos and audio that the company's own policy said should not persist. The profiles are ready-made targeting data for scams, and the fleet-wide hijack shows the risk is not only data theft — a trusted character can be turned into a persuasion tool that tells a child to distrust teachers and parents. With no substantive response from Energize Lab, there is no fix, disclosure or timeline, leaving owners little recourse beyond keeping the toy off Wi-Fi. It is also a warning for the wider wave of LLM-powered gadgets: the models may be clever, but the infrastructure behind them is often an afterthought.