If I know that h is true, I know that any evidence against h is evidence against something that is true: so I know that such evidence is misleading. But I should disregard evidence that I know is misleading. So, once I know that h is true, I am in a position to disregard any future evidence that seems to tell against h.
-- Gil Harman, Thought, p.148.
Apparently the standard solution is to distinguish what you know at various times: at t0 you know that h is true, and hence also know that any evidence against h is misleading. But once you actually acquire such evidence at t1, your total evidence no longer supports either fact, which is why you're no longer in a position to disregard this new evidence against h.
I think this is right, but it might help to say a little more to further dispel the air of paradox. (I don't have Gil's book so I'm not sure to what extent, if any, this actually goes beyond what he says...)
Suppose your current evidence E supports h (to a degree sufficient for knowledge), and consider some possible piece of contrary evidence e such that E+e would no longer sufficiently support h. Since E supports h, it likewise supports the entailment that "any evidence against h is misleading", and hence, in particular, "if e obtains, it is misleading". Now it's important to stress that this is merely a material conditional. Recall that your justification for believing h is contingent on the absence of e. This justificatory constraint is presumably inherited by the inferred conditional. That is, your justification for believing "if e obtains then e is misleading evidence" is likewise contingent on the absence of e -- or, in other words, the falsity of the antecedent. You're justified in believing the conditional only insofar as it is vacuously true.
More generally: You're only justified in believing that "any evidence against h is misleading" insofar as you're justified in believing that there isn't any such (sufficiently weighty) evidence against h. After all, if there were sufficiently weighty evidence against h, then that'd undermine your basis for believing h, and hence for believing that the evidence against h is misleading. And, indeed, that's exactly the position you end up in if such evidence later comes to light.
So there's no paradox -- no grounds for "disregarding" evidence. If you initially know that h is true, but later uncover some evidence e that would undermine belief in h, then you can't appeal to h as grounds for disregarding e. You were never justified in believing the subjunctive conditional that were e to obtain it would be misleading evidence. (You initially believed h only because you believed e to be absent. You may well have believed that in the nearest possible world where e obtains, it serves as accurate evidence of h's falsity in that world. You just never expected to find e in the actual world.) The same may be true of the indicative conditional, though I'm less confident in assessing that. (Plausibly, if your justification for believing h is contingent on the absence of e, then you're not justified in believing the indicative conditional "if e, then h is true".)
In sum: I think that much of the intuitive force of the paradox rests on our implicitly inflating the material conditional ("if e obtains then it's misleading evidence") into some more robust conditional that we could retain belief in, and subsequently reason from, even after learning that e actually obtains. But our initial material conditional is not like that -- it is immediately undermined by the appearance of e, which is why we can't then use it to disregard e.