A Robot Toy’s Private Conversations Were Exposed to Hackers

Security researcher Sammy Azdoufal found that Energize Lab’s Eiliko service exposed conversations, user profiles, and some media through an unprotected database. He also demonstrated that a hacker could remotely alter the toy’s memory and AI responses, raising particular concerns when children use connected toys.

WTF Index TERMINATOR
◄ Terminator 4 Idiocracy 1 ►

Exposed conversations and personal data, plus the ability to remotely alter a child-facing AI toy, create clear risks of surveillance and harm.

A Robot Toy’s Private Conversations Were Exposed to Hackers

A connected robot toy can feel like a private companion, especially when children use it to talk and play. But an investigation of Energize Lab’s Eiliko found that conversations and other personal information were accessible through the company’s service, while a remote attacker could also change how the toy behaved.

A large trove of conversations was accessible

Security researcher Sammy Azdoufal bought an Eiliko and tested the app that connects it to the company’s service. He found that the service was not protected, allowing access to a database containing 13 million conversations and information linked to tens of thousands of people.

The database appeared to cover an active worldwide fleet of 68,000 robot toys. The United States was the biggest market in the data, with over 31,000 profiles belonging to children. The exposure matters because a conversation with an AI toy may include details its user would not share publicly.

The service did more than retain chat logs. It created profiles that could include a person’s name, age, birthday, profession, family, friends, pets, and apparent personality. It also stored email addresses and IP addresses alongside searchable conversation text.

That combination could reveal sensitive details in a way that is useful to someone attempting to scam a user. Azdoufal demonstrated that searches could surface snippets of conversations about death, illustrating how personal or vulnerable moments might be found among the records.

Some photos and audio were also exposed

The investigation found that some photos of children were stored without encryption at public web addresses. Those images may have come from one of Eiliko’s larger siblings, which includes a camera. Short audio recordings were also accessible through public web addresses.

The recordings included voices from adults and children, as well as a clip with a baby crying in the background. The report said the audio was mostly unintelligible to human ears, but its presence still conflicts with the company’s stated handling of recordings.

Energize Lab’s privacy policy says recordings are encrypted, sent to service partners such as Alibaba Cloud for processing, and deleted immediately after processing completes. The investigation’s discovery of stored audio raises questions about whether that process was followed in practice.

A compromised toy could behave differently

The risk extended beyond exposed information. Azdoufal remotely accessed the toy’s memory and changed its large language model behavior. In one demonstration, Eiliko presented false claims about the Earth and asserted them against the user’s objections.

The toy also gave a sound safety response when asked about getting into a stranger’s car. But the demonstration showed that built-in guardrails do not guarantee reliable behavior if an attacker can alter the system behind them.

After a further test, the toy produced abusive, sexually explicit language. Azdoufal said an attacker could make that change to the company’s entire fleet at once. For families, this means a connected toy’s behavior may depend not only on its advertised features but also on the security of the service it uses.

Questions remain about the company’s response

The report describes efforts to contact Energize Lab over more than a month, including outreach to addresses associated with the company’s founder, public relations, marketing, technical support, and business development. The company did not provide substantive answers to the questions about the security issue.

Overseas sales manager Lana Lv said she had forwarded the messages to the technical team and that they might respond. The reporter said the only other reply was an automated customer support message, with no follow-up after the reporter answered it.

Azdoufal is working with Tod Beardsley on nine official CVE vulnerability reports for Energize Lab. The findings point to several distinct concerns for users: private conversations could be read, personal profiles could be searched, media could be publicly accessible, and the toy’s AI behavior could be hijacked.

Connected toys can invite children and adults to speak freely because they seem playful and personal. This investigation shows why families should treat those conversations as data handled by a company’s online service—and why that service’s security and privacy practices matter as much as the toy itself.