Emergent Collaborative Deliberation in Multi-Model AI Systems: A BFT-Derived Protocol for Epistemic Synthesis

The Consilium Protocol, derived from Byzantine Fault Tolerance, is introduced for structured multi-model AI deliberation, treating inter-model disagreement as an epistemic signal. The study demonstrates that cognitive personas determine epistemic behavior and that RLHF alignment training creates measurable epistemic blind spots.