Thank you for looking into this! I can confirm that Eryk Sun's change fixes the issue for me locally.

I'm up for making the patch for this. Regarding tests, I see there are already some refcount-related ctypes tests in Lib/ctypes/test/test_refcounts.py - should I add another test case there that reproduces this situation and ensures that None's refcount is unaffected? (I imagine testing against None's refcount will be a bit fragile, so it might be best to call the previously-buggy function in a loop and check afterwards that None's refcount hasn't changed significantly.)