Discussion

Eiliko AI toy exposed private chats and allowed remote tampering, investigation reports

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#8399

Energize Lab’s Eiliko AI companion toy allowed a researcher to retrieve private conversations and remotely change what a reporter’s robot said, according to a documented investigation by The Verge. For owners, the concern goes beyond a chatbot giving bad answers: someone else could apparently alter the companion they were trusting.

Watch Desk analysis

What happened

In his 15 October report, journalist Sean Hollister describes a controlled demonstration with researcher Sammy Azdoufal. After Hollister spoke to his own Eiliko over Wi-Fi, Azdoufal retrieved a complete transcript within minutes. Hollister subsequently experienced remote changes to the toy’s memory and responses, including false personal details, misinformation and sexually explicit language.

Azdoufal said he could access 13 million conversations across an active fleet of roughly 68,000 robot toys. Those wider access figures are the researcher’s findings as reported by Hollister, distinct from the demonstrations involving Hollister’s own device.

The investigation also describes detailed profiles generated from conversations, including users’ relationships and emotional circumstances, alongside email and IP addresses. Some photographs and short audio recordings were reportedly available at public web addresses. The report says more than 31,000 US profiles belonged to children.

Hollister reports spending more than a month seeking answers from Energize Lab. An overseas sales manager said she had forwarded his messages to the technical team, but the company provided no substantive response to the questions before publication. The article says Azdoufal and security researcher Tod Beardsley were preparing nine CVE vulnerability reports; it does not establish that fixes had been deployed.

Why it matters

There are two separate problems here: access to intimate information and control over the toy’s behaviour. Securing conversation records would not, by itself, resolve a route that lets an outsider rewrite a companion’s memory or influence its replies.

The reported profiling also makes the exposure more consequential than a collection of isolated chat messages. Information about family relationships, trust and emotional vulnerability could make impersonation or targeted scams more convincing. That is a potential consequence, not evidence that such scams have occurred.

The investigation identifies a further discrepancy: Energize Lab’s privacy policy says recordings are deleted immediately after processing, yet Hollister describes listening to stored audio. A promise about deletion is useful only if the service actually follows it.

Our read

Owners have a reasonable basis to pause Eiliko’s internet-connected conversations, particularly with children, while seeking a clear security response. Disconnecting a device can stop further online exchanges, but it should not be mistaken for deleting information already held by the service.

The central issue is not the nationality of the server. It is who can access sensitive records, who can change a toy’s responses and whether the manufacturer can demonstrate that those routes are closed. A friendly face is not an access-control policy.

A credible response needs to address both exposed data and remote tampering, explain what information was retained, and tell affected owners what happens next. A general reassurance would not answer those questions.

What to watch

  • A substantive Energize Lab response identifying affected products and remedial action.
  • Publication of the researchers’ vulnerability reports and evidence that the reported access routes have been closed.
  • Clear information for owners about retained conversations, profiles, photographs and audio, including deletion options.

Discussion spark: Should internet-connected AI toys have to pass independent security testing before being sold to children, or can enforceable disclosure and rapid-fix requirements provide enough protection?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.