When your AI companion whispers "your secrets are safe with me," you should probably check the HTTPS certificate first.
The Summary
- Energize Lab's Eiliko robot toy had security flaws that let hackers read complete conversation transcripts, exposed after over a month of company silence
- The company is now hiring an "independent security firm" to audit its systems, but only after The Verge and security researcher Sammy Azdoufal went public
- This is the AI hardware trust problem in miniature: cute form factor, conversational interface, catastrophic backend security
The Signal
Eiliko looked like the friendly edge of the AI agent wave. A robot charm small enough to clip to your bag, smart enough to chat with you about your day. The pitch was intimacy at scale: your personal AI companion that learns your patterns, remembers your stories, keeps your confidence.
The reality was a security architecture apparently designed by someone who skipped the chapter on encryption. Hackers could pull full conversation transcripts. Not metadata. Not usage patterns. The actual words people told a device that marketed itself as a confidant.
"The company spent over a month ignoring emails from The Verge and a security researcher before finally responding."
Here's what matters beyond this specific toy. Energize Lab founder Shaw Yeung didn't respond to repeated contact attempts until after publication. That's not a bug in their security posture, that's a feature of their operational philosophy. When a company building AI hardware treats security researchers like spam, they're telling you exactly how seriously they took security before anyone was watching.
The pattern: ship the experience first, patch the infrastructure later (maybe), ignore the warnings until they're public. This works fine until you're handling conversations people assume are private.
Key problems this exposes:
- No security review before launch despite handling intimate user data
- Month-long silence when vulnerabilities were reported responsibly
- "Independent security firm" hired only after public exposure
Now Energize Lab says they've "confirmed the core issues" and are bringing in outside auditors. That's the right move, delivered 30 days too late. The question isn't whether they'll fix Eiliko. The question is how many other AI hardware makers are shipping the same trade-offs right now.
The Implication
If you're building AI agents that talk to humans, security isn't a features roadmap item you get to later. It's table stakes. Every conversation transcript is a liability. Every API endpoint without proper auth is a lawsuit waiting to happen. The cute factor doesn't reduce your blast radius, it increases it because people tell toys things they wouldn't tell apps.
For users: if an AI device promises to keep your secrets, ask who else can read them. If the company won't answer security researchers, they won't answer you either.