I’m going to take a different road here. Honestly I wonder if the risk/reward of doing that isn’t more important.
For anyone reachable by the USA it would be incredibly risky to hack the FBI. They aren’t going to pay you the millions you might get from hitting a commercial entity and they’re sure going to come find you to save face.
For someone in an adversaries geo space you don’t have to worry about the risk to much and the reward might be more worthwhile.
Then again, they might also already have that information 20 times over.
With that being said, I wonder more if the reason isn’t necessarily just AI but also because these systems have rarely been tested (by experienced threats) before.