AI Governance and Risk Management: Frameworks, Standards, and Taxonomies#

As AI systems become deeply embedded in critical infrastructure, healthcare, finance, and defense, the need for structured governance frameworks has never been more urgent. This guide walks through the major frameworks, standards, taxonomies, and practical checklists that security professionals and organizations can use to manage AI risk responsibly.

Risk Management Frameworks#

NIST AI Risk Management Framework#

The NIST AI Risk Management Framework is the cornerstone of US federal AI governance. Published by the National Institute of Standards and Technology, it provides a structured approach to identifying, assessing, and managing AI risks throughout the system lifecycle.

Key characteristics of the NIST AI RMF:

Voluntary but influential — While not legally mandated, it shapes federal procurement requirements and is referenced by multiple regulatory proposals
Lifecycle approach — Covers risks from data collection through model training, deployment, and decommissioning
Trustworthy AI characteristics — Defines seven properties: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed

ISO/IEC Standards#

The International Organization for Standardization has published several critical AI standards:

ISO/IEC 42001 — The first international standard for an AI Management System (AIMS). Similar in structure to ISO 27001 for information security, it provides a certifiable framework for organizations to manage AI risks systematically.
ISO/IEC 23894:2023 — Guidance on AI risk management, complementing the 42001 standard with practical risk identification and treatment strategies.
ISO/IEC 22989:2022 — Establishes foundational AI concepts and terminology, providing a common language for cross-organizational AI risk discussions.

Industry Frameworks#

Framework	Publisher	Focus
Secure AI Framework (SAIF)	Google	End-to-end AI security across the model lifecycle
ENISA Multilayer Framework	ENISA	Cybersecurity practices specific to AI within the EU
AI Maturity Assessment	OWASP	Assessing organizational readiness for secure AI adoption
CSA AI Model Risk Framework	CSA	Risk management for AI models in cloud environments
CSA Maestro	CSA	Threat modeling specifically for agentic AI systems

Standards and Verification#

OWASP AI Security Verification Standard#

The OWASP AISVS provides a comprehensive set of security requirements and test cases for AI systems. Think of it as the AI equivalent of the ASVS (Application Security Verification Standard) — it gives auditors and developers a clear checklist of what secure AI looks like.

Agent Discovery and Observability#

As AI agents proliferate, two new standards address their management:

OWASP Agent Name Service (ANS) — A standardized approach for secure AI agent discovery, analogous to DNS for traditional network services. Prevents agent spoofing and unauthorized access.
OWASP Agent Observability Standard — Defines how AI agents should expose telemetry, metrics, and audit trails for security monitoring and incident response.

AI Verify#

AI Verify is a Singapore government-backed testing framework and toolkit. It allows organizations to verify AI system properties against governance frameworks through automated testing, providing evidence of compliance and trustworthiness.

Taxonomies and Risk Databases#

Understanding the threat landscape requires a common language. These resources provide that foundation:

Attack Taxonomies#

NIST AI 100-2e2023 — The authoritative taxonomy for adversarial machine learning, defining attack types, threat models, and defense categories.
MITRE ATLAS — The Adversarial Threat Landscape for AI Systems. Maps AI-specific attack techniques in a format familiar to anyone who uses MITRE ATT&CK.
AVIDML — An open-source taxonomy for AI vulnerability disclosure, enabling standardized reporting of AI security issues.
Arcanum Prompt Injection Taxonomy — A comprehensive classification system for prompt injection attacks, covering attack intents, techniques, evasions, and input vectors.

Risk Databases#

MIT AI Risk Repository — A comprehensive database of AI risks categorized by domain, severity, and mitigability.
AI Incident Database — A crowdsourced repository of real-world AI failures and harms, invaluable for understanding how AI risks materialize in practice.
CSA LLM Threats Taxonomy — Cloud Security Alliance’s taxonomy specific to LLM threats.

Scoring and Assessment#

OWASP AI Vulnerability Scoring System (AIVSS) — Extends CVSS concepts to AI-specific risk dimensions, providing a standardized way to score and compare AI vulnerabilities.

Practical Checklists and Guidance#

OWASP has produced an extensive library of practical guidance documents:

LLM Cybersecurity and Governance Checklist — A point-by-point checklist for securing LLM applications in production.
Agentic AI Threats and Mitigations — Addresses the unique risks of autonomous AI agents.
Securing Agentic AI Applications — A guide to building and deploying secure agentic systems.
GenAI Incident Response Guide — How to handle security incidents involving generative AI systems.
LLM and GenAI Data Security Best Practices — Guidance on protecting training data, prompts, and model outputs.
AI Security Solutions Landscape — A mapping of available tools and solutions to specific AI security challenges.
GenAI Red Teaming Guide — Methodology for systematically testing AI systems for vulnerabilities.

Implementing AI Governance in Your Organization#

Based on these frameworks, here is a practical approach:

Phase 1: Assessment#

Use the OWASP AI Maturity Assessment to understand your current state
Map your AI systems using the NIST AI RMF trustworthiness characteristics
Identify gaps using the AISVS verification standard

Phase 2: Framework Selection#

If you need certification, implement ISO/IEC 42001
If you are cloud-heavy, adopt the CSA AI Model Risk Framework
If you are building agents, use the CSA Maestro framework

Phase 3: Ongoing Management#

Subscribe to MITRE ATLAS and AI Incident Database updates
Use the OWASP checklists for every new AI deployment
Establish an AI security review process modeled on the GenAI Red Teaming Guide

Key Takeaways#

AI governance is no longer optional — regulations like the EU AI Act and federal procurement requirements are making it mandatory
The combination of NIST, ISO, and OWASP frameworks provides comprehensive coverage from risk management to technical implementation
Taxonomies like MITRE ATLAS and the Arcanum Prompt Injection Taxonomy give teams a shared language for discussing threats
The OWASP practical checklists are the quickest path to improving your organization’s AI security posture today

Newsletter