We tried it. DOGE was a complete failure of tech startup idiots asking models questions as if they were oracles and blindly trusting them to make subjective decisions about which Congressionally-created programs we should kill. Some of the tech were smart enough to realize how much damage they did and got out quickly.

Adversarial inputs at the level of the government somewhat fascinate me.

(aeons old blackmail tactic requires more work than the ability to select which of two data batches is more aligned with adversarial behavior)