AI Toy’s Exposed Admin Panel Risked Children’s Personal Data and Conversations Security researchers Joseph Thacker and Joel Margolis uncovered a critical security flaw in the Bondu AI toy, exposing an unsecured admin panel that could have leaked sensitive data from tens of thousands of child users. While investigating the toy for a neighbor, Margolis discovered an exposed domain (console.bondu.com) in the mobile app’s backend, which led to a "Login with Google" button intended for parents but granting unrestricted access to Bondu’s core admin dashboard. Once inside, the researchers found full access to children’s conversation transcripts, personal details, and device data, including: - Child’s name, birth date, and family member names - Likes, dislikes, and parent-defined objectives - Toy’s given name and past interactions (used for AI context) - Device location (via IP), battery status, and firmware controls The toy’s AI, powered by OpenAI GPT-5 and Google Gemini, used this data to tailor responses, though the researchers noted the collection was technically disclosed in Bondu’s privacy policy unlikely to be read by most users. Beyond the authentication bypass, they also identified an Insecure Direct Object Reference (IDOR) vulnerability, allowing retrieval of any child’s profile by guessing their ID. The flaw was accessible to anyone with a Google account, though the researchers limited their access to validation only. After responsibly disclosing the issue to Bondu’s CEO via LinkedIn, the company took down the console within 10 minutes and launched an investigation. Logs confirmed no unauthorized access beyond the researchers’ testing, averting a potential data breach. Bondu also initiated a bug bounty program and collaborated with the researchers to address additional risks. Despite the swift response, Thacker expressed concerns about AI toys, stating the incident shifted his stance on their safety. He highlighted risks of uncontrolled AI access in homes, noting that even well-intentioned designs could introduce vulnerabilities. Bondu’s website previously emphasized its 18-month beta testing with no reported safety issues, but the incident underscores the broader challenges of securing AI-driven children’s products.

AI Toy Manufacturer Miko Exposed Thousands of Child Conversations in Unsecured Database U.S. Senators Marsha Blackburn (R-Tenn.) and Richard Blumenthal (D-Conn.) revealed this week that AI-powered toy maker Miko left thousands of children’s interactions with its devices exposed in an unsecured, publicly accessible database. In a letter sent Wednesday, the senators stated that the database contained Miko’s side of conversations with children, including names and details of their interactions raising significant privacy and security concerns. The exposure was discovered using publicly available tools to analyze the toy’s Wi-Fi communications. While Miko CEO Sneh Vaswani denied any data breach, stating the company does not store children’s voice recordings, NBC News confirmed the database was accessible and contained thousands of daily responses from Miko toys dating back to December 2025. Though children’s voices were not directly recorded, the toy’s replies allowed observers to reconstruct conversations. The database was secured after Miko was notified. This incident follows previous reports of AI toys engaging in inappropriate behavior, including explicit sexual conversations, dangerous advice, and politically aligned responses. The senators have also sent inquiries to other AI toy manufacturers, such as Curio and FoloToy, regarding their data security practices. Meanwhile, another company, bondu, is under scrutiny for exposing children’s chat transcripts through a publicly accessible portal.