I believe (with only moderate confidence, I could easily be convinced otherwise by good evidence) that the network if you get 51% of delegates into a hostile state, then this requires the userbase of Bitshares to hard fork to fix the situation. A hostile delegate includes one that is accepting bribes in order to collaborate with the other hostile delegates.
Ultimately, a blockchain derives value from the size of its network. Such an attack would split the network into the two forks, and users would be forced to choose between them. It should be fairly clear what has occurred, and almost all of the users should choose the chain without the hostile delegates. If people do not support one chain unanimously, then the value of the network will be split into the two forks depending on the number of users who support each one.
Obtaining a delegate spot is relatively difficult, and requires one to build a reputation. Maintaining a delegate position is also difficult and requires creating value for bitshares, so if the attack is not executed very quickly, there are ongoing costs.
https://bitsharestalk.org/index.php?topic=13864.msg180306#msg180306If many delegates are replaced in a short period of time, this is a massive warning flag to everyone to be extra aware of the possibility of attack.
The cost to a delegate who engages in hostile activity (either directly or taking bribes to collaborate), is to potentially lose their delegate position. Either the attack fails, and those who tried are voted out. Or the attack succeeds, the userbase is upset at the results, and hardforks back to a good state and removes the offending delegates.
From a game theory perspective, the major thing working in our favor is that bad actions result in loss of reputation, and it is difficult (expensive) to build reputation.