I believe there’s reason to think that Eliezer never intended “Sufficiently optimized agents appear coherent” to have an airtight argument and be universally true.
On the Arbital version of the page (but not the GreaterWrong version you linked to) [ETA: I just realized that you did link to the Arbital version, but I was viewing it on GW] one can see that Eliezer assigned 85% probability to the claim (though it’s not clear if the uncertainty is more like “I tried to make an airtight universal argument, but it might be wrong” or more like “I tried to show that this will happen in most cases, but there are also cases where I don’t think it will happen”).
On the Arbital version of the page (but not the GreaterWrong version you linked to) [ETA: I just realized that you did link to the Arbital version, but I was viewing it on GW] one can see that Eliezer assigned 85% probability to the claim (though it’s not clear if the uncertainty is more like “I tried to make an airtight universal argument, but it might be wrong” or more like “I tried to show that this will happen in most cases, but there are also cases where I don’t think it will happen”).