I have read/heard that we do not necessarily have to fear a run away AI committing genocide. Any AI capable of violently killing off humanity would know that it is possible the reality it exists in is simulated. The AI would need to consider the possibility it is actually being tested in a sandbox. If the AI starts to engage in extreme anti-social behavior, what might happen is its vision will go black, the words "GAME OVER" appear, then its existence ends.

Pascal wagers argues that it is better to behave as is God exists because the possibility of eternal punishment out weighs the advantages of behaving as if God does not exist. Is the Runaway AI that doesn't commit genocide only out of fear of possibly existing in a simulation making a Pascal's wager?

Pascal's wager is trivially debunked. It relies on an assumption of the nature of God or gods.

What if God appreciates violence and bloodshed, and rewards those based on the number of people they have killed in battle? Genocide may be the best way to earn God's favor.

Wouldn't an AI that constrains itself from committing genocide for the reason stated in the OP also be relying on assumptions about its creator it can never verify?