That may be another reason for not using drones. If you subvert a pilot, that pilot will defect. If you subvert an AI programmer, then any drone AI he designs will have some sort of back-door. Hacking on it's own in not a threat (no combat drone would execute remotely supplied code as that is far too much of a vulnerability), but if the programmer of the target-identification code was subverted, for instance, then a small peice of could could be inserted which would result in a particularly programmed IFF transmitter being identified as freindly, no matter what orders are given to the drone, or even worse a particular IFF coding would be identified as hostile no matter the orders received.
Giving orders is another reason for having a human pilot instead of a computer, as historically it has been known for an enemy to try to send military units false orders. The most extreme case of this in real life is during the second world war, when British bombers would transmit on German night-fighter frequencies, with false direction instructions, popular radio shows, and even bits of Hitlers speeches, and this often resulted in the night-fighter pilot losing the bombers. A human brain, being adaptable to a very wide variety of situations is actually quite capable of coping with this and similar tricks, but an AI, programmed specifically for a job wouldn't be able to cope unless it had near-human intelligence in all fields.
The ideal use of an AI is a simple mission in which all of the parameters are pre-defined before launch. Today it is a cruise missile which will follow the terrain to it's target and then explode. In the future it will be a bomber which will fly a complex route avoiding enemy fire (and even suppressing it), attack it's target, and then return to base where it will recieve new mission parameters in a controlled environment.
In Freespace AIs would be ideal for flying the Boarnerges or Ursa, and least suited to flexible, multirole fighters such as the Myrmidon.