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.

It's not a binary. If you expect to receive 90 meritorious complaints and 10 bad ones, a review procedure with no false negatives and 10% false positives will give you the wrong outcome 1% of the time. You might be OK with that depending on the cost of a false positive. But if the rise of AI means you get 100 meritorious complaints and 100 bad ones, and you apply the same process, now your processing costs have doubled and your error rate has quintupled.

But if it requires human vetting, and AI massively increases the volume of requests, then it also massively increases the effort to properly vet those requests.