Eiliko robot maker Energize Lab pledges independent security review and pulls memory features after fleet-wide flaws are exposed

In an October statement from its CTO, Energize Lab said it had rotated credentials, closed external backend access and disabled the remote interface that could change robot responses. When the email arrived, The Verge and researcher Sammy Azdoufal still found photos and audio at unprotected public addresses, and Azdoufal still had full SSH access to the robots. [1]

The flaws were serious and affected the whole platform. They include nine CVEs, among them a path traversal that reads secrets, AI model administration with no authentication, admin APIs with no authorization checks, default SSH credentials and unsigned firmware downloaded over HTTP. Together they affect roughly 68,000 robots and millions of stored conversations. [2] [3] [4]

The cloud long-term memory feature became an attack surface. Editable user profiles were fed back as trusted context, so prompt injections persisted across conversations. Endpoints also skipped ownership checks, so one account could read other families' records. A test robot was made to spread falsehoods, invent personal details and produce sexually explicit, coercive dialogue. [3] [6] [8]

Disclosure stalled for weeks. According to the advisory, the company never answered the researcher. A September 14 retest found 42 of 44 issues still exploitable, a 72-hour deadline expired on September 18, and details went public on October 15. [3]