The problem then doesn't have to do with AI specifically, it requires human intervention and vetting. If the problem is "someone submitted an AI-written request that got rubberstamped", the problem is the stamp, not the AI.

Interesting, your observation on societal structures changing is very close to mine regarding laws: A lot of "laws" were created with the assumption that they would be applied as warranted by circumstances, and I think universal AI surveillance and perfect enforcement of laws will be a huge problem in the coming age.