Trust, but Verify

We demonstrate that in a cluster of mostly honest nodes, we can detect nodes that run unauthorized or incorrect large language models (LLMs) through the social consensus of their peers. We discuss an intersubjective validation system to encourage honest behavior from LLM nodes.