>How do you test and monitor outbound access against program that adapts itself to get around things? Current generation LLMs don't do that. Also you can just

>How do you test and monitor outbound access against program that adapts itself to get around things? Current generation LLMs don't do that. Also you can just prompt it to not hack systems on the internet and the model follows that instruction. You can also add a simple classifier to internet requests. The classifier model can obviously be as intelligent as your hypothetical rogue model. Nothing you said makes any sense in the context of LLMs or the current news cycle.