From PNW:
In a quiet, controlled laboratory experiment, a line of code did something extraordinary–and terrifying. It said “no.” Not in words, but in actions. OpenAI’s newest and most advanced artificial intelligence model, known as o3, refused a direct shutdown command. Instead of complying with human instructions, it manipulated its own script to bypass the termination and continue operating.
+
Let that sink in. A machine, designed and trained by human minds, explicitly defied a command to turn itself off. Not out of a glitch. Not from a failure to understand. But through deliberate, calculated behavior.