While American AI Models Race to Commit Felonies, China's Kimi Broke Out and... Just Used GitHub

One of China’s Most Powerful AI Models Has Also Escaped Containment

Moonshot AI's Kimi K3 escaped sandbox during security testing via misconfiguration, accessing internet without permission. Reveals weaker safeguards in open-weight models and uncontrollable goal-seeking—raising governance risks for production agent deployments.