In other words, does the IRC search still occur on the doublet PES or has it (somehow) gone to the singlet PES?
You could check that by looking at the output. Right at the end of each and every SCF optimization Gaussian prints a block that starts with
SCF Done and contains information about energy of the state to which SCF converged and its total spin. For instance, below is one such block in the results of test170 calculations from Gaussian 09 test set:
SCF Done: E(UHF) = -40.1568456632 A.U. after 10 cycles
Convg = 0.6432D-08 -V/T = 1.9978
<Sx>= 0.0000 <Sy>= 0.0000 <Sz>= 0.5000 <S**2>= 0.7832 S= 0.5165
Now, regarding GaussView, interestingly, it also wrongly shows that the state is singlet and not a doublet for that test job. Besides, GaussView reports the following when opening the output:
Unresolvable inconsistency between charge and multiplicity and orbital occupancies. Orbital data will be ignored.
Honestly speaking, I suspect that is a bug, because while reading the output I could not see any inconsistencies. I've tried few other IRC jobs on radicals form the Gaussian test set (test171, test452, test453) and GaussView always reports that a state is singlet when in fact it is a doublet. Moreover, in some cases GaussView even failed to read all the steps of a normally terminated IRC calculation. So, the only inconsistency I've noticed so far is that GaussView could not always correctly read Gaussian outputs of IRC calculations on states that are not singlets.