I've seen this AI amassing fleet bug in a few multiplayer games now. Our set up was: 2 human players teamed against 3 AI on hard (2 aggressor + 1 eco), medium map, normal fleet size, locked teams, with all 3 AI on the same team.

A single AI player on hard aggressor amasses a fleet one jump away from the target player's planet. It then jumps in to attack. If you kill its cap ship(s) it retreats back to that one jump, even if your fleet is smaller than the AI's, or if you retreat. It then waits to rebuild the cap ship, continually amassing at that one jump away from my planet. What's more, if its cap ship is otherwise occupied (e.g. fighting another player, or defending against pirates), it will continue to amass there without attacking, and without sending the fleet anywhere else -- that fleet just sits there until it decides to attack my planet again.

There were several occasions where the AI attacked without a cap, killed my planet, colonized it for themselves, then went back to the same spot to amass their force further without pressing the attack.

We were consistently sending the pirates at this particular AI, and they had a second fleet (another 30 ships) defending the pirate attacks. The first fleet doing the amassing was often, but not always, one jump away from where the pirate attack was occurring. Note that the AI had an overwhelming advantage over the pirates and did not need their first, amassing fleet to rejoin, and they did not send that first fleet back -- the first fleet amassing was essentially idle.

The AI does the above regardless of whether or not it has an overwhelming advantage in the attack on my planets. I've seen a fleet of 50 ships including siege retreat and amass repeatedly for over 3 hours, despite my defending force of 15 ships (weaker) jumping in and out of the planet in question. During one 6 hour game, that one AI player could have cleaned half of my planets anytime in the last 4 hours of the game, yet it continued to hit one of my outer planets, then retreat to one jump away in order to amass its force again. I saw the AI repeatedly scout my adjacent planets to the one being attacked -- these adjacent planets had only minor defenses and the AI's 50 ship fleet could have walked all over them.

Note that the AI was not doing the following:
- Retreating to defend against another player (other player was fighting a different AI)
- Retreating to defend against pirates (the AI in question had a second fleet defending)
- Retreating purely because of lost cap ships (sometimes they attacked without a cap)
- Retreating due to having a weaker force (in every case, the AI was stronger than the force I had defending)

The only way we were able to break the above amass/attack/retreat cycle was to attack the AI's home planet -- it then stopped amassing or attacking my planets and came back to defend.

I realize that programming AI is extremely difficult and I would hate to try to reproduce this just based on my description above. What I would like to see looked at is the conditions under which the AI presses an attack, particularly when the AI has scouted and found that their forces are superior to the ones nearby.

We've found that due to the above AI limitation/bug and numerous others that hard AI is only hard because it's efficient -- it is ineffective at any form of strategy. There are numerous other posts bragging about beating many hard AI -- these AI bugs are probably a big reason why. (It's not clear if those people boasting are aware of how locked teams works.)

Also please fix the AI's fleet composition.

A big thanks to the AI developer(s) for all their hard work! What you've achieved thus far is really awesome!