Actually, they even use hypothetical variables to “predict” past temperatures and never get it right. They brag if they get closer to the actual than they have been, or, as my late boss, M. Stanton Evans put it, “so they’re getting better at predicting the past.”
That's called "testing" or "ground truth". Start a model with 1980 data and see how well its results compare with 2010 data. For example.
If the models they claim predict conditions for the next century were actually any good, they could be initialized to the conditions of 1918, and then run to construct a prediction of the period 1918-2018, and that prediction would match the actual historical record for that period.
If they could actually do that, they would be shouting it from the rooftops.
But they can’t do that, which means the “models” are utter bunk.