> The reason LLM stays consistent/can be somewhat trusted to stay on rails is that it's aware of both it's context/scope

"It's aware" is a big leap and bold original thought, as awareness is not something it has full stop, try different terminology?