Oh...I love when people exaggerate the truth, especially around AI.
Here's the reality... You can continuously monitor observable events generated by an AI agent (API calls, tool invocations, network activity, shell commands, file access, cloud actions, identity events, etc.).
You cannot continuously determine, with certainty, that an agent has "started hacking." Intent is not directly observable.
But what the fuck would I know, I just worked with AI long before the rest of these putzes found a way to grift...