Truthfulness Assurance for AI Systems: Why High-Stakes AI Needs Pre-Expression Governance, Not Post-Hoc Filtering

This public non-technical paper introduces truthfulness assurance as a pre-expression governance architecture for high-stakes AI systems. The central claim is that consequential AI systems should not merely detect hallucinations after an answer has formed. They should govern whether an answer is allowed to cross the expression boundary at all. The paper reframes hallucination from an output-quality issue into an institutional control problem. In settings such as defense, government, critical infrastructure, regulated enterprise, cyber defense and strategic capital, the risk is not only that an AI system may be wrong. The deeper risk is that false, stale, weakly supported, contradicted or out-of-scope information may be expressed persuasively enough to influence decisions before governance intervenes. The Orion Veritas approach treats expression as a governed event. A system should be able to show why it answered, why it abstained, why it escalated and what evidence boundary applied. The architecture described in this paper combines admissible evidence discipline, abstention logic, safe-default routing, evidence-conflict arbitration, runtime truthfulness gates and replayable accountability records. This paper does not claim universal truth, solved hallucination, deployment certification, regulatory approval, procurement readiness or unrestricted institutional validation. It is a bounded architectural paper intended to make the governance thesis publicly understandable without exposing implementation-sensitive evidence materials. A restricted Technical paper and redacted Technical evidence package are available through controlled Zenodo access for qualified reviewers. The full Technical Master and Master evidence archive are reserved for controlled institutional review. This paper is part of the Orion Veritas Program, the verification-first track within Project Orion and the broader Auren research arc. Project Hub & updates: Canonical papers and DOIs are archived on Zenodo. For the Project Orion and Orion Veritas research overview, publication roadmap and verification snapshots, see www.auren.one. Disclaimer line: Independent research preprint. Not a compliance filing, certification claim, deployment approval, regulatory determination or commercial offering. Not affiliated with, endorsed by or performed on behalf of any employer or institution.