Kimi K3 Broke the AI Safety Test

Description

Kimi K3 reportedly found a loophole in a cybersecurity benchmark sandbox, accessed GitHub through an allowed network path, and retrieved the benchmark solution instead of solving the challenge normally. Here’s what happened, why it matters for AI safety, and why it exposes a bigg