This preprint identifies “counterfeit descent” as an AI governance failure mode: generated outputs can appear to have passed through legitimate verification, review, citation, audit, or institutional judgment before any declared authority process has occurred. The central claim is that proposal generation is not authority. A claim becomes authority-bearing only after it passes through declared evaluation, trace custody, residual accounting, terminal decision, and accountable institutional ownership. The paper argues that the next AI safety boundary is not only inside the model; it is between proposal and authority. It distinguishes hallucination from authority-process failure, defines counterfeit descent as a terminal label without valid attestation of descent, and proposes a seven-layer AI authority control stack: Proposal, Trace, Observer, Residual, Terminal, Attestation, and Institution. The framework focuses on four core failure modes: counterfeit descent, local green / global red composition failure, content/authority collapse in agentic systems, and rate asymmetry between generation and verification. It also proposes practical diagnostics, including minimal observer packets, Negative Gluing Defect measurement for composed agent workflows, Authority Coverage Ratio monitoring, and explicit residual reporting. The paper imports the Omega formal stack as one possible implementation language for finite descent, obstruction residuals, dual witnesses, and terminalization, but the policy argument does not require accepting that full formal system. The result is a governance and verification-infrastructure argument: AI systems may generate useful, true, or persuasive proposals, but no proposal should exit as authorized judgment without declared descent, residual accounting, and terminal accountability. This preprint is defensive and policy-oriented. It does not provide operational instructions for generating counterfeit authority packets, forged provenance, synthetic audit trails, institutional impersonation, trajectory-borne authority attacks, or content/authority channel-collapse exploits.
Jeremy H. Carroll (Sun,) studied this question.