I honestly think the performance of the AI must be situational, in that it depends on the given conditions as to how well it performs. I'm on my second game since Federations doing a run as the Commonwealth of Man. I had a defense pact with another nation early in the game, and he invites me to join a war against another power that was more powerful than each of us separately. I decided to chance it. I made some claims and joined in.
My partner actually participated in the war. By no means was the AI particularly brilliant, but he actually took systems and reclaimed those that were taken by the enemy once the enemy fleet had gone elsewhere. Very rarely did I find one of his fleets with no orders. The enemy fleet also kept switching its attention between myself and my ally depending on which one was making gains. We were able to settle status quo with each of us picking up a few systems.
Then later on, a similar situation against the same enemy, and my ally's performance was more lackluster and confused. He managed to keep his territory defended, but seemed to let several opportunities to take systems go by.
But then again later in the game, when I was head of a large hegemony federation, I declared a vassalization war, and my nearest aliy swarmed all over him. When the war was over, I took the nation as a vassal ... minus two systems that said ally apparently had a claim on. I wondered if that partially drives AI motivation, i.e. what's in it for them. Doesn't explain the reports of failing to defend territory, of course, but it's interesting nonetheless.