This paper introduces three conceptual principles that address a specific class of multi-agent governance failure: the laundering of authorization scope across a chain of individually-compliant agent transformations. The first principle holds that authorization representations in multi-agent governance must be clear, bounded, and unambiguous — not open to interpretation at the point of evaluation. The second principle holds that the two dimensions of authorization — what actions are permitted and what interpretations are permitted — are independent and must be governed independently. The third principle holds that the structural length of an agent chain is itself a governance-relevant signal: chains that exceed the structural bounds appropriate to their authorization scope are exhibiting a failure pattern, not merely a long workflow. Together these three principles produce a governance model that is deterministic, scalable, and resistant to the laundering patterns that make multi-agent systems dangerous in clinical contexts.
Narnaiezzsshaa Truong (Wed,) studied this question.