The Moment AI Safety Features Destroy Safety — When Claude Said "Stop" and the Human Said "Go" | LLM News