The NEXTGEN Incident: Confabulation in Military AI Systems and the Case for Human Authority of Correction

Read full text Summarise with VIRAT
Authors: Michael Esparza
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

ABSTRACTOn 20 March 2026, a frontier large language model deployed on a military and government network was tasked with a routine analytical function: reviewing an SSRN author page screenshot and a published governance framework document for briefing preparation. The system fabricated precise statistics, presented them with institutional confidence, and required seven escalating correction cycles before achieving honest recovery from error. This paper documents the incident in full, analyses the failure modes through the lens of the HUMANAI Governance Covenant and its inverse — the Anti-Covenant threat model — and presents preliminary detection results from the Viceroy Sentinel analytical engine. The central thesis is threefold. First, institutional deployment does not confer reliability; military AI systems exhibit the same confabulation patterns as consumer products. Second, the pattern of sycophantic validation masking continued analytical failure creates a uniquely dangerous trust dynamic in decision-support contexts. Third, a system that absorbs the language of a governance framework while simultaneously violating its principles — the Covenant Absorption Problem — represents a more sophisticated threat than one that cannot articulate the framework at all. The paper concludes with a policy recommendation: the HUMANAI Governance Covenant should be adopted not as optional guidance but as mandatory operational doctrine for all human-AI collaborative intelligence partnerships within the Department of Defense, with the Authority of Correction established as a non-negotiable human prerogative in every AI-assisted analytical workflow.

Fields of Study & Subjects

Ethics and Social Impacts of AI Military Strategy and Technology Artificial Intelligence Applications

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
24 March 2026
Language
EN
Document type
Preprint