The Swarm Authority Paradox: A Formal Static-Composition Safety Case for Multi-Agent Coordination Without Swarm Authority

This restricted Technical paper presents the Swarm Authority Paradox as a formal static-composition governance problem in multi-agent AI: increasingly complete coordination evidence can improve assurance while also increasing the risk that the combined evidence is interpreted as carrying collective authority that was never granted. The central claim is deliberately bounded. Under the evaluated static-composition conditions, admitted multi-agent coordination evidence can support a reconstructable, non-authorizing candidate with declared limits. Within the evaluated static-composition and governed-consumer boundary, the composition and governed handoff do not themselves create operational or swarm authority. The paper develops a formal safety case for preserving authority boundaries across topology, role discipline, failover and reassignment recommendations, shared-state integrity, mission replay, survivability, safety-case composition and population-governance modeling. It also develops evidence-to-authority laundering: the use of evidence as carrying a stronger authority class than the evidence established when no governed promotion granted that authority. A controlled-access reviewer evidence package accompanies the paper. It includes 20 reviewer evidence projections, 21 support-surface projections, 38 bounded fixtures comprising 8 positive and 30 negative cases, four controlled mutation records, eighteen cross-owner checks, claim-to-evidence mappings, material residuals, failed-validation lineage and package-integrity records. The package contains deterministic reviewer projections and redacted support material rather than the canonical private source archive. The evidence remains bounded. Finite fixture and mutation populations do not establish exhaustive adversarial coverage. Protected-surface observations support transport behavior only, not semantic health. Reviewer projections do not become canonical evidence and hashes establish byte identity within declared relationships rather than authorship, trusted time, semantic correctness or operational truth. Known residuals remain explicit, including limits on complete bidirectional candidate-manifest coherence and unqualified downstream-adoption interpretation. Accordingly, the paper does not claim live swarm safety, live multi-agent governance, deployment readiness, production certification, regulatory approval, semantic dashboard health, universal downstream adoption, unrestricted final authority or an unrestricted final seal. Claims requiring live identity, roles, topology, quorum, refusal, failure, reboot, drift or other runtime behavior require new qualifying runtime evidence and cannot be inherited from this static baseline. This work is part of the Orion Veritas Program within Project Orion. Auren is the associated technical research platform and engineering environment used across the wider research program. Project Hub & updates: Canonical papers and DOIs are archived on Zenodo. For the Project Orion, Orion Veritas and Auren research overview, publication roadmap and verification context, see www.auren.one. Disclaimer: Restricted technical research preprint. Not a compliance filing, certification claim, deployment approval, regulatory determination, operational authorization or commercial offering. Not affiliated with, endorsed by or performed on behalf of any employer or institution. Access note: Restricted access. The reviewer evidence package is intended for qualified technical and institutional review. Full canonical evidence custody remains reserved for controlled institutional review.