Researcher Sammy Azdoufal found that Energize Lab's Eilik and Eiliko AI companion robots store their conversations on cloud servers in Beijing with protections missing or incomplete, leaving searchable transcripts and user profiles accessible. The counts differ by how the outlets counted the researcher's findings: Cybernews counts about 8.3 million conversations stored as readable text, while The Verge counts 13 million conversation snippets and an active fleet of about 68,000 robots, with more than 31,000 United States profiles belonging to children. The profiles combine names, ages, email addresses, and IP addresses, and some photos and audio recordings were accessible at public web addresses. The Verge witnessed a live demonstration in which the researcher altered its own reporter's robot.

A flaw allowing anyone to change the AI model controlling a robot without authentication, CVE-2026-105038, is rated 9.8 out of 10 by Cybernews, and the researcher says the fleet could be manipulated at once. The figures above are the researcher's findings as reported by both outlets; Energize Lab gave no substantive response to either outlet, and owners are advised to avoid sensitive conversations and consider disabling cloud chats until the flaws are fixed.

Sources: Cybernews; Malware.news (reporting The Verge)