Your sub-agents should not outright lie to each other ("should" in the predictive, not normative, sense - let me know if it turns out yours do), but they may threaten, negotiate, hide, and be genuinely ignorant about themselves.

I disagree! Finding out when we are lying to our 'selves' about either beliefs or values is one of the most valuable outcomes of this exercise. Lying to our selves so that we may more effectively lie to each other is a core human (homo hipocritus) skill!