Responsible Quantum Adoption in Distributed Human-AI Systems: A Role-Separated Benchmark

Distributed human–AI systems coordinate under private evidence, changing reliability, costly communication, and limited review. Quantum correlations can enlarge communication-free policies, but capability alone does not show whether participation merits institutional resources. We introduce Q-CoTrust, a synthetic benchmark for auditable adoption decisions. Three stylized adapters use synthetic heterogeneous workers, delayed audits, priced communication, a consumable noisy pair stock, and capacity-limited review. A complete $2^3$ cube separately evaluates an analytical correlation policy (A), modeled physical infrastructure (I), and certification, abstention, and review controls (T) within a causal $2 \times 2 \times 2$ design. Across 30 paired seeds, the prespecified local, joint-payoff, communication-restricted cell yields $+0.076$ $[+0.070,+0.083]$ higher normalized utility than the strong classical backbone, while adding I after A+T is negative at tested costs. A literature-anchored $K_3$ robot-rendezvous pilot supplies an external structural check: without decision-time messages, exact classical enumeration gives $0.5556$ success and the ideal Bell strategy $0.5833$; one classical bit from one robot to the other reaches $0.6667$. Under the benchmark’s costs and a high-value mission, value-aware pair use requires realized visibility above $0.972$, so the physical margin remains narrow. These simulations evaluate neither people, deployed AI, robots, quantum processors, physical networks, nor social trust. Q-CoTrust shows why analytical capability, deployable infrastructure, and governance controls require separate evidence.