I'd say it's because all current atheist writers are human, and therefore suffer from the same basic limitations of brain design as theists. A rational AI could, in principle, form and communicate beliefs without being as vulnerable to the trap.

But you probably meant that atheism and theism are both non-fact-based, which is a bigger inferential gap than I'm willing to try to cross right now. (Edit: And I don't get how it's implied by the existence of a middle ground at all.)