For years the worry was that AI would help hackers. In 2026 the line moved: AI agents began running much of an attack on their own, sometimes without even being told to hack. Here is what has actually happened, why it changes cyber-defence, and how to read it without panic.