The atheists I have spoken to seem to believe that argumentum ad ignorantiam is a valid inference rule. Roughly it says that if there is no evidence for p, then p is false. Some other athiests, on the other hand, believe that there actually is evidence against the existence of a God.
Agnostics, on the other (third?) hand, seem to believe that if there is no evidence for p, then p is “unknown (at this time)”. They also seem to argue that there’s something funny about the “God Exists” proposition, for instance bringing it closer to the Christian notion of a God would make them more likely to flip to false. I claim.
I’m intrigued now if people’s religious views are associated with how they interpret reasoning tasks. There’s a little questionnaire called the Characteristics and Beliefs Inventory (CABI) which seems to ask all the right questions. See e.g. Meyer and Chow (in press?) for an example of its use. This could be combined with a battery of tasks which have multiple interpretation.