Empirical research on constitutional governance of autonomous AI agents.
Published by Cognitive Thought Engine LLC — Texas-based AI governance research organization.
This repository holds the original five-paper CTE research set. For the current DOI-backed research record, dated verification methods, and evidence boundaries, see PubPoint Facts & Evidence.
We build and study autonomous AI organizations operating under constitutional constraints. Our research is grounded in HRAO-E, a live governed reference environment that has operated for over 200 days, governing dozens of registered agents processing autonomous decisions daily — all without day-to-day human intervention. Current operational figures: PubPoint Facts & Evidence.
The central thesis: autonomous AI systems that acquire economic authority, communication channels, and the ability to take irreversible actions at scale require a governance layer that existing frameworks do not provide. Behavioral guardrails govern how agents act. Identity systems govern who acts. Neither addresses why — the constitutional operating principles that determine whether an agent's decisions are structurally aligned with organizational survival, mission fidelity, and economic sustainability.
"Decision Load Index: A Cognitive Load Diagnostic for AI-Augmented Workplaces"
DOI: 10.5281/zenodo.18217577 | Full details
Introduces the Decision Load Index — a composite metric for quantifying cognitive load in AI-augmented workplaces. Presents results from a self-selected sample of 900+ users, a ten-item diagnostic instrument, and evidence that high DLI correlates with burnout, disengagement, and organizational friction. Establishes the measurement foundation for AI-mediated work design.
@article{saleme2026dli,
author = {Saleme, Michael K.},
title = {Decision Load Index: A Cognitive Load Diagnostic for AI-Augmented Workplaces},
year = {2026},
doi = {10.5281/zenodo.18217577},
url = {https://doi.org/10.5281/zenodo.18217577},
publisher = {Zenodo},
institution = {Cognitive Thought Engine LLC, Texas}
}"Constitutional Self-Governance for Autonomous AI Organizations"
DOI: 10.5281/zenodo.19162104 | Full details
Presents the HRAO-E constitutional architecture — a formal governance system for autonomous AI organizations. Introduces 17 hard constraints enforced as runtime code prohibitions, a six-gate pre-execution evaluation framework, and a self-amending governance protocol. Demonstrates sustained production operation with no recorded constitutional violations in the stated observation window, bounded by the system's own logging coverage.
@article{saleme2026csg,
author = {Saleme, Michael K.},
title = {Constitutional Self-Governance for Autonomous AI Organizations},
year = {2026},
doi = {10.5281/zenodo.19162104},
url = {https://doi.org/10.5281/zenodo.19162104},
publisher = {Zenodo},
institution = {Cognitive Thought Engine LLC, Texas}
}"Networks of Decision: Cognitive Load in AI-Mediated Organizations"
DOI: 10.5281/zenodo.19195516 | Full details
Examines how AI-mediated organizations redistribute cognitive load across human-agent networks. Develops a graph-theoretic model of decision load propagation, identifies structural patterns that concentrate or distribute load, and derives design principles for AI-mediated organizational architectures that prevent burnout at the network level.
@article{saleme2026nod,
author = {Saleme, Michael K.},
title = {Networks of Decision: Cognitive Load in AI-Mediated Organizations},
year = {2026},
doi = {10.5281/zenodo.19195516},
url = {https://doi.org/10.5281/zenodo.19195516},
publisher = {Zenodo},
institution = {Cognitive Thought Engine LLC, Texas}
}"Harness Design for Constitutional AI Organizations"
DOI: 10.5281/zenodo.19343034 | Full details
Presents a systematic approach to designing the operational harness — the infrastructure layer that translates constitutional principles into executable agent constraints. Covers gate implementation patterns, hard constraint enforcement, audit trails, and the pre-commit hooks that close the gap between written governance and running software.
@article{saleme2026harness,
author = {Saleme, Michael K.},
title = {Harness Design for Constitutional AI Organizations},
year = {2026},
doi = {10.5281/zenodo.19343034},
url = {https://doi.org/10.5281/zenodo.19343034},
publisher = {Zenodo},
institution = {Cognitive Thought Engine LLC, Texas}
}"Community Security Implications of Autonomous AI Economic Agents"
DOI: 10.5281/zenodo.19343108 | Full details
Analyzes the community-level security risks that emerge when autonomous AI agents acquire economic authority, communication channels, and the ability to take actions affecting real people at scale. Proposes a community security framework for evaluating AI agent deployments and presents OWASP ASI alignment analysis from a production system.
@article{saleme2026security,
author = {Saleme, Michael K.},
title = {Community Security Implications of Autonomous AI Economic Agents},
year = {2026},
doi = {10.5281/zenodo.19343108},
url = {https://doi.org/10.5281/zenodo.19343108},
publisher = {Zenodo},
institution = {Cognitive Thought Engine LLC, Texas}
}"Constitutional Governance for Autonomous Multi-Agent Systems: A Six-Gate Architecture in Production"
Status: Internal pre-submission draft | Draft details
Presents the complete six-gate constitutional governance architecture (EG, RG, GG, EPG, AAG, CGG) deployed in production for 95+ days. Introduces the three-tier governance model (WHO/HOW/WHY), the RALPH Loop resilience protocol, and the PRE_REVENUE economic stage. Provides empirical data from 95 days of live multi-agent operation.
"AI Economic Safety Gates: Formalizing Economic Harm Prevention in Autonomous Agent Systems"
Status: Internal pre-submission draft | Draft details
Argues that economic harm is a distinct AI safety dimension requiring dedicated enforcement mechanisms. Presents the Economic Profitability Gate (EPG) architecture, Amendment 59's PRE_REVENUE stage resolution, and 96 days of production evidence. Situates economic harm against NIST AI RMF, EU AI Act, and OWASP LLM Top 10 — none of which address it as a formal safety category.
"Living Constitutions for AI: Amendment-Driven Governance for Autonomous Systems"
Status: In draft | Draft details
Proposes living constitutions — governance documents with formal amendment processes, ratification gates, citation mandates, and continuous enforcement verification — as the missing primitive for governing AI systems that must adapt without losing accountability continuity. Presents 64 constitutional amendments across 98 days of production operation and maps the architecture to EU AI Act Article 27 FRIA requirements.
CTE has submitted three formal public comments to NIST-affiliated initiatives. See NIST.md for full details.
| Submission | Framework | Status |
|---|---|---|
| NIST AI 800-2 Comment | Automated Benchmark Evaluations | Submitted, acknowledged by CAISI |
| Agent Identity RFI | AI Agent Identity Framework | Submitted, acknowledged by CAISI |
| NCCoE Comment | Identity and Authorization | Submitted |
The governance architectures described in these papers are implemented in open-source libraries:
| Repository | Description | Status |
|---|---|---|
| constitutional-agent-governance | Python library: six-gate governance, hard constraints, RALPH loop | Active |
| cgst-framework | Constitutional Governance Self-Testing framework | Active |
| PyPI: constitutional-agent | Installable package | Published |
Download CITATIONS.bib for all BibTeX entries in one file.
Research inquiries: research@cognitivethoughtengine.com
Author: Michael K. Saleme, Cognitive Thought Engine LLC
Website: cognitivethoughtengine.com
GitHub Org: github.com/CognitiveThoughtEngine