This is also BS. The AI does not weigh any option. It sends it sends all its ships in one fleet to your homeworld or sector.
I'm gonna try this one more time. Please read what I write and not what you think I write before you throw around the "BS" moniker - it's really rather tedious.
Making a strategic AI for a game like Stellaris is VERY different from making a strategic AI for eg EU4. In EU4 during wars you have fairly limited options in targets, paths and enemies that make it relatively easy to code an AI that can weigh a number of parameters and arrive at a decent course of action.
In Stellaris you have a much more fluid battlefield. Someone else compared it to fighting on the seas instead of inland and that's a good illustration. You do not have strong borders, fortifications and simple paths through a few provinces. You have a scenario where at any time you can get ambushed by the enemy, and where your targets are nearly limitless as are your liabilities.
Because of this - which humans have a hard time dealing with as well - the team behind Stellaris have coded the AI to be "safe". That means the allied AI always try to stack up and enemy AI go for what it perceives the highest value target in a bunch of more or less equal targets - the home worlds.
The reason the AI is so bad is because coding an AI for a dynamic and fluid battlefield is EXTREMELY hard. Why do you think most recent 4x games in space - including the remake of MOO - have ONLY hyperlanes? Because with hyperlanes you can severely limit options and thus improve AI weightings.
Stellaris has ambitiously decided not only to not use only hyperlanes, but not even just to use Warp exclusively which would be the second easiest AI to make, but Wormholes as well - simultaneously! This is unprecedented in 4x and will demand more from the AI than in any other such game. If they can make it work well it will be truly impressive.