In Church's "First Anonymous Referee Report on Fitch's 'A Definition of Value'" from 1945 we find not only the first formulation of the proof today known as the knowability paradox, but we find the first proposed solution. Church rejects any principle stating that propositional attitudes necessitate (other) propositional attitudes. For instance, Church rejects plausible closure principles for belief. He explains,
To be sure, one who believes a proposition without believing its more obvious logical consequences is a fool; but it is an empirical fact that there are fools. It is even possible there might be so great a fool as to believe the conjunction of two propositions without believing either of the two propositions; at least, an empirical law to the contrary would seem to be open to doubt. On this ground it is empirically possible that a might believe k' at time t without believing k at time t (although k' is a conjunction one of whose terms is k.
Church here denies that belief is closed under conjunction elimination. The context reveals that he is also denying that knowledge is so closed. More generally the idea is that if a propositional attitude is the result of another, it is so contingently. Notice that we have today departed dramatically from Church's thought. In all of the literature on the knowability paradox, for instance, it is granted that it is at least metaphysically necessary that one knows the conjuncts of known conjunctions. An exeception is found in a position articulated in Kelp and Pritchard's very interesting forthcoming " Anti-realism, Factivity and Fitch".
To pick another example,
...there is no valid law of psychology according to which anything whatsoever about my desires may be inferred from the fact that I know so-and-so.
It should be noted that Church (in the second referee report) allows Russell's theory of types to avoid such obvious counterexamples as "Necessarily, if a knows that she desires that p then a desires that p". Accordingly, the approach blocks the instance of the factivity of knowledge that is employed in the knowability paradox---viz., K(p & ~Kp) => (p & ~Kp). In this regard Church forshadow's Bernie Linsky's independent thoughts on the matter, forthcoming as "Logical Types in Arguments about Knowability and Belief".
In sum, Church's position is that his knowability paradox is indeed an invalid proof. On his view knowledge is neither closed under conjunction elimination nor unrestrictedly factive.