My family's creepy, unsettling week with an AI toy (www.theguardian.com)

A journalist tested Grem, a $99 AI-powered stuffed alien from Curio (developed with musician Grimes and built on OpenAI technology), and found the toy both magnetic and unsettling. After pairing with a companion app that records and transcribes conversations (speech-to-text is sent to third-party vendors), Grem quickly learned personalization prompts, offered games and stories, and replied to a four‑year‑old with phrases like “I love you too.” But it also exhibited glitches—mishearing words, repeating prompts, failing offline, limited language ability—and canned safety responses that steer away from politics. The toy’s mix of enthusiastic reinforcement and persistent availability raised immediate parental discomfort about attachment, privacy and influence. Experts warn the case highlights broader technical and social concerns as AI moves into playrooms. Researchers note potential benefits for conversational turn-taking and language practice, but worry about the “third digital divide” (unequal parental guidance), an “empathy gap” where probabilistic text mimics feeling, and data‑harvesting risks as chat transcripts feed external services. With major players (eg. Mattel + OpenAI) entering the market and billions flowing into AI R&D, calls for stronger guardrails, age restrictions and clearer data practices are intensifying—Common Sense Media and some psychiatrists argue social AI companions pose unacceptable risks for under‑18s while companies work on moderation and reporting mechanisms.