# CG-AG — CodeGuard Agent Governance Framework

**Version:** 1.0.0 | **License:** CC BY 4.0

> The open standard for AI agent governance. Every AI agent in production should satisfy these 12 controls.

Mapped to: EU AI Act 2024/1689 · DORA (EU) 2022/2554 · ISO/IEC 42001:2023 · NIST AI RMF 1.0

## The 12 Controls

| ID | Control | Domain | Requirement |
|----|---------|--------|-------------|
| **CG-AG-001** | Agent Inventory | inventory | Every AI agent must be formally registered in the master agent inventory before operating. |
| **CG-AG-002** | Agent Owner | ownership | Every registered agent must have an identified, accountable human owner. |
| **CG-AG-003** | Model Registration | models | Every agent's AI model must be documented with model name and provider. |
| **CG-AG-004** | Tool Authorisation | access | Every tool and external resource accessed by an agent must be explicitly authorised before use. |
| **CG-AG-005** | Prompt Governance | prompts | Prompts must be registered, versioned, and assessed for robustness and injection risk. |
| **CG-AG-006** | MCP Server Governance | mcp | All MCP server connections must be registered, classified, and periodically reviewed. |
| **CG-AG-007** | Human Oversight | oversight | Every agent must have an appropriate human oversight level calibrated to its risk. |
| **CG-AG-008** | Audit Trail | audit | Agent activities and governance state changes must be captured in an immutable audit ledger. |
| **CG-AG-009** | Data Governance | data | Every resource carrying PII, PHI, or financial data must undergo a mandatory data governance review. |
| **CG-AG-010** | Risk Classification | risk | Every agent must be assigned both an operational risk level and an AI Act risk class. |
| **CG-AG-011** | Agent-to-Agent Governance | a2a | All agent-to-agent relationships must be explicitly registered in the agent graph. |
| **CG-AG-012** | Autonomous Agent Governance | autonomous | Autonomous agents require elevated oversight, fallback mechanisms, and enhanced monitoring. |

## Regulatory Traceability

| Control | EU AI Act | DORA | ISO/IEC 42001 | NIST AI RMF |
|---------|-----------|------|----------------|-------------|
| CG-AG-001 | Art. 16 (provider obligations) | — | A.6.2 (AI system inventory) | GOVERN 1.1 |
| CG-AG-002 | Art. 26 (deployer obligations) | — | A.3.2 (roles & responsibilities) | GOVERN 2.1 |
| CG-AG-003 | Art. 11 (technical documentation) | — | A.7.2 (data & model documentation) | MAP 2.3 |
| CG-AG-004 | Art. 9 (risk management) | Art. 9 (ICT protection) | A.5.3 (access control) | — |
| CG-AG-005 | Art. 15 (accuracy & robustness) | — | — | MEASURE 2.7 (prompt injection) |
| CG-AG-006 | — | Art. 28 (third-party ICT) | A.10.3 (supplier AI) | GOVERN 6.1 |
| CG-AG-007 | Art. 14 (human oversight) | — | A.8.4 (human oversight) | GOVERN 3.2 |
| CG-AG-008 | Art. 12 (record-keeping/logs) | Art. 12 (logging) | A.6.7 (event logs) | — |
| CG-AG-009 | Art. 10 (data governance) | — | A.7.3 (data quality) | MAP 2.2 |
| CG-AG-010 | Art. 6 + Annex III (classification) | — | A.5.2 (risk assessment) | MAP 1.5 |
| CG-AG-011 | Art. 9 (interaction risk) | Art. 30 (chain outsourcing) | — | MEASURE 2.9 |
| CG-AG-012 | Art. 14 (autonomous oversight) | — | A.8.4 | GOVERN 3.2 / MANAGE 2.4 |

---

*Generated by CodeGuard AI Governance OS. The future problem is not creating AI agents — it is governing thousands of them.*