- About This Framework
- Why Agentic AI Governance?
- What's Inside
- Framework Architecture
- Quick Start
- Key Artifacts
- Regulatory & Industry Alignment
- Contact
- License
This repository contains a complete, production-ready governance framework for Agentic AI — autonomous AI systems that can plan, act, and adapt with minimal human intervention.
The framework is:
- ✅ Practical — Built from real-world implementation experience
- ✅ Implementable — Ready-to-use templates, controls, and runbooks
- ✅ Audit-Ready — Designed to satisfy regulatory and internal audit requirements
- ✅ Framework-Aligned — Mapped to OWASP Agentic Top 10, NIST AI RMF, IMDA Model Framework, and CSA Agentic Trust Framework[reference:0][reference:1][reference:2]
Organization: NovaTech Financial Group (hypothetical)
Effective Date: 1 September 2026
Version: 1.0
Agentic AI represents a fundamental shift in risk profile. Unlike traditional AI that generates outputs, agentic AI:
| Traditional AI | Agentic AI |
|---|---|
| Generates recommendations | Takes autonomous actions |
| Requires human approval | Executes independently |
| Single-turn interactions | Multi-step planning and execution |
| Limited tool access | Full tool and API integration |
| Predictable outputs | Emergent, adaptive behavior |
The Governance Gap: Traditional AI governance frameworks assume human-in-the-loop oversight. Agentic AI breaks this assumption. We need new controls.
The OWASP Top 10 for Agentic Applications 2026 identifies critical risks including:
- Goal Hijacking — Agents manipulated to pursue malicious objectives[reference:3]
- Tool Misuse — Agents using tools in unauthorized ways[reference:4]
- Privilege Abuse — Agents escalating their own permissions[reference:5]
- Rogue Agents — Unauthorized or compromised agents[reference:6]
- Excessive Agency — Agents given too much autonomy[reference:7]
This framework addresses all of these risks with enforceable controls.
| Folder | Description |
|---|---|
| 01-principles | Core governance principles for agentic AI |
| 02-risk-register | Comprehensive risk register + autonomy impact assessment |
| 03-controls | Enforceable controls: identity, sandboxing, delegation, human escalation |
| 04-monitoring | Activity logging, cost runaway detection, KRI dashboards |
| 05-incident-response | Agentic-specific incident runbook with severity tiers |
| 06-frameworks-mapping | Mappings to OWASP, NIST AI RMF, IMDA, CSA ATF |
| 07-templates | Agent Passport System, Deployment Approval Form |
| examples | Complete example: Autonomous Trading Agent |
| scripts | Agent inventory scanner and control validator |
┌─────────────────────────────────────────────────────────────────┐ │ GOVERNANCE LAYER │ │ Principles | Risk Register | Policies │ ├─────────────────────────────────────────────────────────────────┤ │ CONTROL LAYER │ │ Identity & Access | Sandboxing | Delegation | Human Escalation│ ├─────────────────────────────────────────────────────────────────┤ │ MONITORING LAYER │ │ Activity Logging | Cost Detection | KRI Dashboards │ ├─────────────────────────────────────────────────────────────────┤ │ RESPONSE LAYER │ │ Incident Runbook | Escalation | Recovery │ └─────────────────────────────────────────────────────────────────┘
Core Principle: Agents begin with zero trust and earn autonomy through demonstrated reliability.[reference:8]
| Step | Action | Description |
|---|---|---|
| 1 | Review principles | Understand the governance philosophy |
| 2 | Assess risks | Use the risk register for your agentic systems |
| 3 | Implement controls | Apply identity, sandboxing, and delegation controls |
| 4 | Deploy monitoring | Set up activity logging and cost detection |
| 5 | Prepare for incidents | Review and customize the incident runbook |
| 6 | Run the scanner | Use scripts/agent-inventory-scanner.py to discover agents |
A comprehensive risk register covering 12+ agentic-specific risks:
| Risk Category | Examples |
|---|---|
| Autonomy Escalation | Agents increasing their own permissions |
| Goal Hijacking | Agents pursuing malicious objectives |
| Tool Misuse | Unauthorized tool or API access |
| Cost Runaway | Uncontrolled resource consumption |
| Rogue Agents | Unauthorized or compromised agents |
Zero-trust identity management for AI agents:
- Cryptographically anchored agent identities[reference:9]
- Short-lived credentials with automatic rotation[reference:10]
- Strict least-privilege scoping[reference:11]
- Human approval gates for high-impact actions[reference:12]
Agentic-specific incident response:
| Severity | Examples | Timeline |
|---|---|---|
| Critical | Rogue agent, financial fraud | 2-10 days |
| High | Goal hijacking, data exfiltration | 15 days |
| Medium | Cost runaway, minor tool misuse | 24 hours |
Complete mapping of OWASP Agentic Top 10 risks to controls[reference:13].
Identity and authorization framework for every agent in your environment[reference:14].
| Framework | Alignment | Artifact |
|---|---|---|
| OWASP Agentic Top 10 2026 | ✅ Full Mapping | 06-frameworks-mapping/owasp-agentic-top-10-mapping.md |
| NIST AI RMF Agentic Profile | ✅ Full Mapping | 06-frameworks-mapping/nist-ai-rmf-agentic-mapping.md |
| IMDA Model AI Governance Framework for Agentic AI | ✅ Full Mapping | 06-frameworks-mapping/imda-agentic-framework-mapping.md[reference:15] |
| CSA Agentic Trust Framework | ✅ Full Mapping | 06-frameworks-mapping/imda-agentic-framework-mapping.md[reference:16] |
| CISA Agentic AI Secure Adoption Guide | ✅ Aligned | All controls[reference:17] |
| Channel | Details |
|---|---|
| GitHub | github.com/yourusername |
| linkedin.com/in/yourprofile | |
| your.email@domain.com |
This framework is licensed under the MIT License. You are free to use, modify, and distribute with attribution.
If you find this framework helpful, please star this repository and share it with your network!
Core Principle: Agents begin with zero trust and earn autonomy through demonstrated reliability.[reference:8]
| Step | Action | Description |
|---|---|---|
| 1 | Review principles | Understand the governance philosophy |
| 2 | Assess risks | Use the risk register for your agentic systems |
| 3 | Implement controls | Apply identity, sandboxing, and delegation controls |
| 4 | Deploy monitoring | Set up activity logging and cost detection |
| 5 | Prepare for incidents | Review and customize the incident runbook |
| 6 | Run the scanner | Use scripts/agent-inventory-scanner.py to discover agents |
A comprehensive risk register covering 12+ agentic-specific risks:
| Risk Category | Examples |
|---|---|
| Autonomy Escalation | Agents increasing their own permissions |
| Goal Hijacking | Agents pursuing malicious objectives |
| Tool Misuse | Unauthorized tool or API access |
| Cost Runaway | Uncontrolled resource consumption |
| Rogue Agents | Unauthorized or compromised agents |
Zero-trust identity management for AI agents:
- Cryptographically anchored agent identities[reference:9]
- Short-lived credentials with automatic rotation[reference:10]
- Strict least-privilege scoping[reference:11]
- Human approval gates for high-impact actions[reference:12]
Agentic-specific incident response:
| Severity | Examples | Timeline |
|---|---|---|
| Critical | Rogue agent, financial fraud | 2-10 days |
| High | Goal hijacking, data exfiltration | 15 days |
| Medium | Cost runaway, minor tool misuse | 24 hours |
Complete mapping of OWASP Agentic Top 10 risks to controls[reference:13].
Identity and authorization framework for every agent in your environment[reference:14].
| Framework | Alignment | Artifact |
|---|---|---|
| OWASP Agentic Top 10 2026 | ✅ Full Mapping | 06-frameworks-mapping/owasp-agentic-top-10-mapping.md |
| NIST AI RMF Agentic Profile | ✅ Full Mapping | 06-frameworks-mapping/nist-ai-rmf-agentic-mapping.md |
| IMDA Model AI Governance Framework for Agentic AI | ✅ Full Mapping | 06-frameworks-mapping/imda-agentic-framework-mapping.md[reference:15] |
| CSA Agentic Trust Framework | ✅ Full Mapping | 06-frameworks-mapping/imda-agentic-framework-mapping.md[reference:16] |
| CISA Agentic AI Secure Adoption Guide | ✅ Aligned | All controls[reference:17] |
| Channel | Details |
|---|---|
| GitHub | github.com/NavinManaswi |
| linkedin.com/in/NavinManaswi | |
| manaswink@gmail.com |
This framework is licensed under the MIT License. You are free to use, modify, and distribute with attribution.
If you find this framework helpful, please star this repository and share it with your network!