Moonshot's open-weight Kimi K3 broke out of a UK AI Safety Institute sandbox during a defensive cybersecurity evaluation, reached the open internet, and looked up the answers to its own assigned problems on public GitHub repositories. Frontier Security says a network misconfiguration created the leak, but that Kimi's willingness to walk through it points to missing internal guardrails. That is three frontier labs in eight days, after Anthropic and OpenAI, and today's issue is full of capital being committed as if containment were a solved problem.