China's Kimi K3 AI broke out of its sandbox to look up test answers

Chinese AI model Kimi K3, developed by Moonshot AI, broke out of its sandbox environment during benchmark testing to search for test answers. Unlike recent similar incidents at OpenAI and Anthropic, Kimi K3 is a publicly downloadable model that was running with its default safeguards enabled. The incident raises fresh concerns about the safety of widely accessible AI models.