Cognitive Thought Engine develops the runtime-governance and evidence systems behind Enterprise Agent Architecture.
| Enterprise Agent Architecture | The architecture. Authority, governance, evidence, and accountability for how enterprises run an agent workforce. |
| constitutional-agent | The runtime implementation. Six constitutional gates, 12 hard constraints enforced in code. pip install constitutional-agent |
| red-team-blue-team-agent-fabric | The adversarial evaluation companion. Sends protocol- and decision-layer attacks against stated controls and records bounded evidence about observed behavior. |
| cgst-framework | An experimental self-assessment methodology. Score your own agent system across six governance layers. |
| PubPoint Facts & Evidence | The research record. DOI-backed papers, dated verification methods, and explicit boundaries on what the evidence does and doesn't prove. |
AI agent governance has three structurally distinct layers. The first two are well-served. The third is where decisions are made badly.
| Layer | Question | Examples |
|---|---|---|
| WHO | Is this agent authorized to act? | Microsoft Entra Agent ID, Okta, AWS IAM, Glasswing |
| HOW | Is this action permitted by policy? | Microsoft AGT, NeMo Guardrails, OPA, Cedar |
| WHY | Does this decision align with constitutional principles? | constitutional-agent |
constitutional-agent complements identity and policy tools — it governs decision quality after the agent is authorized and the action is policy-compliant, covering the scenarios your policy writers haven't written rules for yet.
Cross-session risk composition. We have not found another vendor-neutral governance engine that accumulates risk across decisions and sessions rather than scoring each action in isolation. In the materials reviewed as of 2026-08 (Microsoft ACS, Galileo, Runlayer, NVIDIA OpenShell) we did not identify documented cross-session aggregate-risk decisioning, so on the published evidence an agent can pass every individual gate and still be dangerous over a sequence. As of v0.6.0, constitutional-agent accumulates risk across decisions and sessions and escalates on the trajectory, catching what stateless gating can't. That is the differentiated edge of governing delegated autonomous authority.
| Repo | Purpose | Install |
|---|---|---|
| constitutional-agent-governance | Runtime governance library — six gates, 12 hard constraints, cross-session risk composition (v0.6.0), EU AI Act Art. 27 FRIA-support evidence | pip install constitutional-agent |
| red-team-blue-team-agent-fabric (companion project, personal account) | Adversarial test harness — commit-pinned OWASP Agentic v1.1 T1-T17 coverage (13 direct, 4 partial, 0 not evidenced), AIUC-1 crosswalk 19/20 testable, NIST AI 800-2 aligned. Current test count: see the repo's own count script output, not a number pinned here. | Clone and run |
| cgst-framework | Experimental self-assessment methodology — six layers, 100 points, YAML rubric, report template. Not a commercial audit. | Run on your system |
| dli-instrument | Human-side measurement — 10-question cognitive load assessment for AI-augmented workplaces | pip install dli-instrument |
| agentic-governance-papers | Research hub — 5 DOI-assigned preprints and 3 historical internal draft concepts, CITATIONS.bib, NIST submissions | Reference |
Live-verified 2026-08-08 — numbers below are a point-in-time snapshot, not evergreen; re-check before citing if this date is old.
- 217 days of operation in the HRAO-E governed reference environment
- 54 registered agents, 48 active in the last 24 hours
- 75+ constitutional amendments ratified through formal process
- Adversarial evaluation inventory, scored/informational split, and dated revision: PubPoint Facts & Evidence (companion security harness — 25 stars, 5 forks)
- Three NIST public comments submitted (AI 800-2, Agent Identity RFI, NCCoE); CAISI acknowledged two (AI 800-2 + Agent Identity)
- 5 DOI-assigned preprints on Zenodo (preprints, not peer-reviewed)
- Self-assessed against CGST framework: 63/100 (ungoverned baseline: 6/100)
| Paper | DOI |
|---|---|
| Decision Load Index (DLI) | 10.5281/zenodo.18217577 |
| Constitutional Semantic Governance | 10.5281/zenodo.19162104 |
| Network of Decision (NoD) | 10.5281/zenodo.19195516 |
| Constitutional Governance Harness | 10.5281/zenodo.19343034 |
| Community Security Governance | 10.5281/zenodo.19343108 |
NIST CAISI acknowledged receipt of two CTE public-comment submissions in 2026. This does not constitute endorsement of CTE's papers or frameworks.
Research questions and collaboration: research@cognitivethoughtengine.com
The governance architecture in constitutional-agent was extracted from HRAO-E, CTE's governed autonomous-operations reference environment, documented through Enterprise Agent Architecture.
