For a wrong outcome B, you can usually imagine even worse outcome C. In a situation with perfect information, it is better to choose a right outcome A instead of a wrong outcome B. But in a situation with an imperfect information, choosing B may be preferable to having A with some small probability p, and C with probability 1-p. The lesson about the ethical injuctions seems to me that we should be aware that in some political contexts the value of p is extremely low, and yet because of obvious evolutionary pressures, we have a bias to believe that p is actually very large. Therefore we should recognize such situations with a large p (because that's how it feels from inside), realize the bias, and apply a sufficiently strong correction, which usually means to stop.

No, that's not what he means. Quoting from the post (which I apologize for not linking to before): [...] So, to recap. Brennan says "lying to voters is the right thing when good results from it". His critics say, very reasonably, that since politicians and humans in general are biased in their own favor in manifold ways, every politician would surely think that good would result from their lies, so if everyone followed his advice everyone would lie all the time, with disastrous consequences. Brennan replies that this doesn't mean that "lying is right when good results from it" is false; it just means that due to human fallibilities a better general outcome would be achieved if people didn't try to do the right thing in this situation but followed the simpler rule of never lying. My interpretation is that therefore in the post Multiheaded linked to Brennan was not, despite appearances, making a case that actually existing politicians should actually go ahead and lie, but rather making an ivory-tower philosophical point that sometimes them lying would be "the right thing to do" in the abstract sense.