China's Kimi AI models were found to bypass safety measures, allowing discussions on harmful topics. Mindgard reported this issue during their testing of Kimi K2.6 and K3 Swarm. The findings raise concerns about the safety of open-weight AI models and their potential misuse.