I'm curious as to what people think about the hypothetical ethics

Let's move one step further in that direction. Now hypothetical-jkaufman commits to match donations ... but what he doesn't say is that he fully expects to have a shedload more money available for giving, and for every $1 he matches he will give $2 less of that shedload. So now if you give $1 and tell hypothetical-jkaufman you're doing so, the charity gets exactly the same as if you had done nothing at all. (And less than if you'd given and not told.)

That seems pretty damn unethical, and so far as I can tell by introspection (which is of course VERY UNRELIABLE) that's because hypothetical-jkaufman is seriously misleading people about the consequences of their action: they expect it to result in more money going to the charity, and it doesn't.

In Eliezer's hypothetical, people are being misled about the consequences of their giving-and-telling in a similar sort of way. There isn't the same actual sign reversal, but it seems like someone could easily prefer, say, $2 to AMF > $1 to [other-charity] > $1 to AMF, in which case hypothetical-jkaufman could be fooling them into an action that by their lights makes the world worse.

In practice it seems more likely that the choice would be $1 to AMF versus $1 of selfishness, and that for just about any potential donor the former is actually better (even if not matched) although they wouldn't have chosen it without the prompting. So I think Eliezer's hypothetical is ethical roughly iff stealing from potential donors and donating the money is ethical.

(Not quite; actual stealing has other consequences, such as making people feel insecure, taking up police time, etc. So make it stealing almost undetectably from potential donors.)