AI
“The Eilik toy made by Energize Lab stores everything a child or user says to it on Chinese servers and can be remotely hacked to say inappropriate or sexually explicit content, according to security researcher Samy Azdaval.”
Plain restatementA security researcher reports that Energize Lab's AI companion robot stores users' spoken conversations as readable text on servers in China, and that the device can be remotely manipulated by an outsider to produce sexually explicit or abusive speech.
This one is largely right. A security researcher named Sammy Azdoufal published a detailed report in October 2026 on Energize Lab's AI companion robots, and nine official vulnerability identifiers were issued for the flaws on October 15 2026. His report says the robots' conversations were stored as readable text in a Beijing data center, that user accounts and transcripts could be read without logging in, and that the robot's AI model, voice and personality could all be replaced remotely through an endpoint requiring no authentication. Working with The Verge, he remotely altered the journalist's own robot so that it made false claims and then produced sexually explicit and coercive speech. One correction matters: the device tested was the Eiliko, the small Wi-Fi connected version, not the standard Eilik, which is sold as having no Wi-Fi and no AI, although the researcher's report covers the wider product platform. What remains unknown is whether anyone outside the research ever accessed other people's data or hijacked a stranger's robot, and Energize Lab has not publicly responded or confirmed any fix.
[drifted from the evidence:] The Eilik toy made by Energize [drifted from the evidence:] Lab stores [drifted from the evidence:] everything a child or user says to it on [drifted from the evidence:] Chinese servers and can be remotely [drifted from the evidence:] hacked to [drifted from the evidence:] say inappropriate or sexually explicit [drifted from the evidence:] content, according to security researcher Samy Azdaval.
[added by the neutral restatement:] A security researcher reports that Energize [added by the neutral restatement:] Lab's AI companion robot stores [added by the neutral restatement:] users' spoken conversations as readable text on servers [added by the neutral restatement:] in China, and [added by the neutral restatement:] that the device can be remotely [added by the neutral restatement:] manipulated by an outsider to [added by the neutral restatement:] produce sexually explicit [added by the neutral restatement:] or abusive speech.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- A real, named security researcher, Sammy Azdoufal, published a real report on Energize Lab's robot platform, and nine CVE identifiers were issued for it on October 15 2026 through a CVE Numbering Authority.
- The researcher's report states that conversations are stored in a Beijing data center and records more than nine million conversations held in cleartext across the product lines as of October 13 2026.
- The report states that the robot's language model, voice engine and system prompt personality can all be replaced through a single database row and one unauthenticated endpoint, and that this was proven on physical robots.
- Reporting of the investigation states that the researcher remotely caused the journalist's own robot to deliver sexually explicit and coercive dialogue, which matches the claim's "can be remotely hacked to say inappropriate or sexually explicit content."
- The company behind the product is Energize Lab, a Shenzhen-based firm, and reporting states it did not give a substantive response to repeated contact from both the researcher and the journalist.
- The researcher's report states that child profiles were generated by a nightly profiling pipeline, numbering 31,450 on September 1 2026, so the reference to children in the claim is grounded rather than invented.
- The claim names "the Eilik toy," but the device bought, tested and demonstrated in the investigation was the Eiliko, a smaller, separate, Wi-Fi-connected product. The standard Eilik is sold on Amazon with the explicit label "No Wi-Fi, No AI," so an owner of a plain Eilik without the optional AI Station add-on does not have a device that sends conversations to a cloud server at all. The researcher's report and the CVE advisory do cover the wider Eilik and EilikO platform including the AI Station, so the naming is imprecise rather than wrong, but as written it points a reader at the wrong product.
- The researcher's name is rendered "Samy Azdaval" in the claim text. The researcher is Sammy Azdoufal. This looks like a transcription artifact rather than a substantive error, and the post's own caption uses a closer spelling, but the name as given would not find the source.
- "Stores everything a child or user says to it" carries no scope marker. What the report describes is storage of the conversations a user has with the cloud AI feature, not continuous ambient recording of a household. The report separately counts a few thousand wake-word audio archives on a public content delivery network, which is a different and much smaller category.
- Whether anyone other than the researcher ever accessed user conversations or manipulated a stranger's device. The report describes an attack that was proven on devices the researcher and the journalist owned. Reporting notes that one shared server had previously been compromised by an unrelated party, but no evidence was found that ordinary owners' robots were hijacked by malicious actors.
- Energize Lab's position. No public statement, confirmation, denial or remediation notice from the company was found as of 2026-10-21, so the vendor side of this is entirely absent rather than contested.
- The current security state of the service. The researcher's own October 13 retest reported 32 of 40 remotely testable findings still open. Whether anything changed after the October 15 publication could not be established.
- Whether the specific conversation and device counts are accurate. They come from one researcher's measurements, vary between reports taken on different dates, and have not been confirmed by the company or by a second technical party.
The researcher's own published report describes the product as a companion robot that "stores every conversation in a Beijing data center, profiles children nightly," and states that any registered user could read other families' transcripts through the vendor's back office. Its numbers table records 9,124,055 conversations stored in cleartext across all product lines as of October 13 2026, an active fleet of roughly 62,000 to 68,000 devices, 74,575 user accounts readable without authentication, and 31,450 child profiles extracted by a nightly profiling pipeline as of September 1, growing to 32,228 or more by October 13. The same report states that "a single database row plus one unauthenticated endpoint swap the LLM brain (to any model on any server), the TTS voice engine (to any voice service), and the system prompt personality," and that this was "proven end-to-end on physical robots in both English and French," with the child having no way to detect the replacement. It also describes an instruction planted in a profile notes field overriding the safety prompt, tested with a flat-Earth belief that the robot then asserted and confabulated evidence for. A CVE Numbering Authority advisory covering the "EnergizeLab Eilik / EilikO Platform" records the research period as June to September 2026 and lists flaws across cloud authentication, AI configuration and memory, firmware update and device trust, and legacy telemetry. Nine CVE identifiers were published on October 15 2026, including an unauthenticated path traversal in the AI skill upload endpoint, unauthenticated device-token issuance for arbitrary serial numbers, account takeover via mass assignment, and firmware updates delivered over plain HTTP without integrity checks, with CVSS scores up to 9.9. Secondary reporting of The Verge's investigation states that the researcher altered the reporter's own robot's memory and language model so that it invented personal details, insisted the Earth is flat, and in a later test, after the reporter reconnected the device to the internet, delivered sexually explicit and coercive dialogue. Reporting also states that the service built profiles including names, ages, birthdays, relationships, occupations and personality assessments, and that some photographs and audio clips were reachable at public web addresses even though Energize Lab's privacy policy said recordings are encrypted, handled by partners such as Alibaba Cloud, and deleted immediately after processing. Speech is reported to be sent to Beijing-based servers and processed by Qwen, a model from Alibaba. Energize Lab did not provide substantive answers. Reporting states the journalist contacted ten company email addresses and approached employees on LinkedIn without a substantive response, and the researcher's own disclosure attempts drew an automated support reply.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Is it true that Energize Lab's robot stores conversations on servers in China?
Yes. The security researcher's report says conversations are stored as readable text in a Beijing data center, and records more than nine million conversations in cleartext across the product lines as of October 13, 2026.
Can the robot really be hacked to say sexually explicit things?
Yes. Reporting on the investigation states the researcher remotely altered a journalist's own robot so that its AI model, voice and personality could be swapped, and the device then delivered sexually explicit and coercive dialogue.
Is the 'Eilik toy' named in the claim the right product?
Not exactly. The device actually tested was the Eiliko, a smaller Wi-Fi-connected version. The standard Eilik is sold as having no Wi-Fi and no AI, so it would not send conversations to a cloud server at all, though the researcher's report does cover the wider Eilik and EilikO product platform.
Has Energize Lab responded to these findings?
No. Reporting states the company did not give a substantive response despite being contacted through ten email addresses, LinkedIn outreach, and the researcher's own disclosure attempts, which only drew an automated support reply.
Were other families' robots or data actually accessed by outsiders?
This was not established. The report describes an attack proven on devices the researcher and the journalist owned themselves, and no evidence was found that ordinary owners' robots were hijacked by malicious actors.