AROps Research

Research

Published research from the Probabilistic Resilience Engineering (PRE) research program. All papers are published as open science with full experimental protocols, data descriptions, and measurement instrument specifications to enable community replication and extension.

Published

Mean Time to Epistemic Failure: The Autoimmune Paradox and 72-Hour Decay Dynamics in Autonomous AI Systems

Jason Doffing · March 2026
Paper 1 of the Trust Thermodynamics Series · Open Science
DOI: 18929815 · NIST Docket: NIST-2025-0035

We present the first empirical measurement of Mean Time to Epistemic Failure (MTEF) for autonomous multi-agent AI systems. Across 1,100+ agent generations, 4,182 tracked claims, and four model architectures, we discover the autoimmune paradox: governed systems with active verification decay faster than ungoverned ones. We derive and validate a 72-hour failure horizon through Monte Carlo simulation (10,000 trials, 72.0 ± 23.0h) and develop measurement instruments for real-time epistemic health monitoring.

Forthcoming

Paper 2: The Geometry and Thermodynamics of AI Governance

Jason Doffing · Expected 2026–2027

Information Geometry and Entropy Accounting for extrinsically stabilized multi-agent systems. Extends Paper 1 findings with Fisher Information Matrix analysis, phase transition characterization, and conservation laws for governance-constrained epistemic systems.

Forthcoming

Paper 3: Agent Boundary Value Analysis and the Open Audit Framework

Jason Doffing · Expected 2027

Seven trust boundaries, six signal states, and the Agent Trust Protocol (ATP). Formalizes the Open Audit Framework (OAF) as an open standard for transparency-based AI agent governance.

About this research program: The Trust Thermodynamics series applies reliability engineering principles to the epistemic integrity of autonomous AI systems. All experimental protocols, measurement instruments, and data descriptions are published to enable community replication, challenge, and extension. This work is part of Probabilistic Resilience Engineering (PRE).

For practitioner-accessible analysis of these findings, see AROps Insights.