CAN-ECS: A Syntactic Control Layer for Trust Propagation in Increasingly Autonomous Multi-Agent Systems

Decentralized trust propagation protocols for autonomous multi-agent systems (e.g., DTPP-Ω) constrain which trust state transitions are valid but typically leave the representational medium of inter-agent communication unconstrained. As agent capability increases — including the possibility of gradual behavioral drift in agents operating over long time horizons — unconstrained natural-language communication between agents creates two risks: semantic drift (meaning shifting gradually across repeated interactions without any single invalid message) and an unnecessarily large expressive attack surface (ambiguity, authority mimicry, scope laundering, and related manipulation patterns that a constrained grammar cannot express). This paper presents CAN-ECS (Compact Assertion Notation — Evolving Control System), a design hypothesis for a fixed, versioned, human-and-agent-legible notation grammar intended to function as part of a multi-agent trust protocol's control boundary, not merely as a compression convenience. We present the architecture as nine falsifiable sub-hypotheses (H1–H9), a corrected verification-escalation formalism, a multi-dimensional Sybil-resistant risk-aggregation rule aligned to DTPP-Ω's existing coalition-detection pattern, and an explicit boundary discipline (H9) distinguishing bounded admission-control risk accounting from the global trust scores that decentralized trust protocols typically prohibit. We report this as a design hypothesis at the pre-implementation stage: several structural defects identified during internal review are documented alongside their corrections, and a concrete evaluation plan and falsification conditions are specified for future empirical work.