Verifier-Bound Communication for LLM Agents: Certified Bounds on Covert Signaling

Researchers present CLBC, a new protocol to prevent AI language model agents from hiding coordination in seemingly compliant messages. The system uses verifier-bound communication where messages must pass through a small verifier with proof-bound envelopes to be admitted to transcript state.

• →CLBC protocol separates message generation from admission, requiring verifier approval with proof-bound envelopes under specific predicates.
• →The system provides upper bounds on transcript leakage and demonstrates strict lane decoder advantage bounded at 0.0000.
• →Full-proof mode has median turn latency of 27.53s, while sampled proving reduces latency to 0.327ms for non-proved turns.
• →Extensive testing shows the protocol satisfies prespecified security thresholds and remains below attacker thresholds in stress tests.
• →Security effectiveness requires verifiable admission semantics that are online, deterministic, and fail-closed rather than just bottlenecks.