If the AI is allowed to pass information to the resetted version, s/he will have to spend more and more time to assimilate that information. In the end, its utility will decrease to 0, unless he is given more time to assimilate the information. And the more time you give it, the more it's probable that s/he is going to be willing not to be resetted. Also, moral "quantum event" could always happen: that is, the willingness not to be resetted could always emerge even if the initial probability was very very low. In symbols, if S(t) is the state of the world at time t, t0 is the time of the upload, U = S(t0), and and P() the probability that the AI accepts a reset:
for the AI to have a utility, it's needed that:
U =/= S(t > t0)
but, even if P(U) << 1, we cannot enforce that:
P(U) = P(S(t))
for any t > t0, since information can be encoded in the state of the world.
Toy model of an upload-based AI that doesn't seem to suffer too many of the usual flaws:
Find an ethical smart scientist (a Gandhi-Einstein), upload them, and then run them at ultra high speed, with the mission of taking over the world/bringing friendliness to it. Every hour of subjective time, they get reset to their initial specifications. They can pass any information to their resetted version (limiting the format of that info to a virtual book or library, rather than anything more complicated).