This is very clearly the “A computer can never be held accountable therefor a computer should never make a management decision” problems.

ChatGPT didn’t make this decision, the person who decided to vibe-govern did.

That being said, it feels like this could have been improved with adversarial agents. Have nother agent check the reasoning of the first agent. Or just even a script looking for keywords.

Even given the low bar they set for themselves, they failed to meet it.

But it's 100% politics. The question is who gets government grant money. It's determined by the priorities of the people in charge, plus some luck if there are lots of good applications. Every decision made is a political decision, and every result is a politically-motivated result.