Note that Eliezer does not give a recommendation for solving this problem. Rather, he gives advice for how to get an unbiased risk assessment. But that is only part of the problem.

I would just like to endorse Hal Finney's remark here. I don't know how to solve the general problem here. I am just suggesting, based on the LA-602 and RHIC Review, that having a classified analysis is better ceteris paribus than not having one.