It is sometimes useful not to artificially exclude the middle when using natural language.

In this case, for example, I suspect it's possible to have a level of what we're calling "security" here that is not so high that it precludes updating on evidence (supposing you're correct that too high a level of security leads to the inability to update), while at the same time being high enough to avoid offense (supposing bobneumann is correct that too low a level of security leads to an increased chance of taking offense).

I do agree that keeping your identity small is also helpful, though.