When it comes to Seasonal-to-Interannual forecasting, we have two possible benchmarks: Previous year's seasonal weather. Historical average climate for that season.

Well you could also add a correction for the measured trend over a longer time period. For example, one can observe that temperature has been generally trending upwards since around 1900, i.e. as the Earth has emerged from the Little Ice Age. In making a basic benchmark prediction, it's reasonable to assume that this trend will continue.

This becomes more of an issue for long-range forecasting, because even if the model does not explicitly use the data it is trying to predict, the researchers working on the model are implicitly aware of the information

I agree, and there also is the file drawer problem, i.e. if a simulation doesn't match history it will be quietly discarded. So when you are presented with a simulation which does match history, you don't know how many simulations were discarded to get to that one. So you don't know how impressive it is that the simulation matches history. Until of course you wait for a few years, observe that the simulation diverges wildly, and conclude that its beautiful fit with history is not very impressive at all.