NIST AI RMF
A voluntary structure for governing, mapping, measuring, and managing AI risk across the lifecycle.
Put it to work Build a shared risk language and a use-case risk record.
AI RMF 1.0 is under revision; check NIST for updates.
FreeFIELD GUIDE / AI SECURITY MANAGEMENT
Make AI useful.
Keep it accountable.
A practical route through governance, risk, and security. Pick where your company is today. Take the tools you need for what comes next.
Where are you in your AI journey?
Choose by use case. Different teams can be at different stages.
PATH 01 / Starting out
Give your first pilot a clear purpose.
For teams exploring AI, choosing tools, or experimenting with a few use cases. Start with the benefit, the people affected, and the capacity to operate it safely.
- Choose a measurable benefit
Define the task, who benefits, and what a successful pilot would improve. Compare it with a workable non-AI baseline.
- Name the owner and the boundaries
Agree who approves the use case, which tools and data are permitted, and which decisions require human review.
- Discover what is already in use
Inventory approved and unofficial AI tools, their owners, data exposure, suppliers, and intended uses.
- Check benefits, risks, and resources
Assess foreseeable harms, privacy and security needs, and whether the team has the skills, oversight, and evidence to run the pilot.
- Make stopping part of the plan
Set evaluation criteria, escalation routes, a fallback, and a review date before expanding the pilot.
What informed this path
Original guidance informed by these notebook topics, with official resources linked below. The three paths are an editorial guide, not a formal maturity standard.
- 01 - AIMS Policy Template_non-c.pdf
- 02 - AIMS Scope Statement Template_non-c.pdf
- 03 - AI Asset Inventory Template_non-c.pdf
- 19_-_Industry_AI_Frameworks_and_Standards_Study_Notes.pdf
- 21_-_Developing_AI_Strategies_for_the_Enterprise_Study_Notes.pdf
PATH 02 / In the middle
Turn pilots into dependable workflows.
For teams with AI in day-to-day work. Connect security, evaluation, suppliers, and human oversight to the way each system is actually used.
- Map the whole system
Record model and application versions, data sources, retrieval stores, integrations, owners, and provider/customer responsibilities.
- Test the actual task
Evaluate quality, privacy, relevant bias, robustness, and misuse using representative examples. Retest when models, prompts, data, or workflows change.
- Set enforceable security boundaries
Limit access and tool permissions, isolate untrusted inputs, validate outputs, and protect sensitive data. A system prompt is not an access control.
- Watch for changes and failures
Assign owners and response thresholds for quality, data changes, safety events, costs, and incidents. Exercise rollback and human escalation.
- Review suppliers and releases
Check retention, training use, sub-processors, change notices, and incident support. Tie each release to evaluation evidence and explicit risk acceptance.
What informed this path
Original guidance informed by these notebook topics, with official resources linked below. The three paths are an editorial guide, not a formal maturity standard.
- 04 - AI Risk Assessment & Treatment Template_non-c.pdf
- 05 - AI Supplier & Third-Party Management Template_non-c.pdf
- 07 - AI Change Management Template_non-c.pdf
- 08 - AI Model Lifecycle Management Template_non-c.pdf
- 09 - AI Data Management & Quality Template_non-c.pdf
- 53_-_Continuous_Monitoring_for_AI_Systems_Study_Notes.pdf
PATH 03 / Advanced stage
Scale the evidence along with the AI.
For organizations running AI across products or business units. Make assurance repeatable, challenge your controls, and use the results to improve them.
- Connect controls to evidence
Map applicable framework requirements to owners, system scope, test results, and review dates. Automate evidence collection where it remains reliable.
- Challenge the controls independently
Use risk-led red teaming, control testing, and independent reviews. Validate relevant failure scenarios and track fixes through retesting.
- Review risk at portfolio level
Track concentration in vendors and models, common data dependencies, and incident trends alongside business outcomes. Set escalation and exception expiry rules.
- Bound autonomy when using agents
Use scoped identities, least-privilege tools, external policy enforcement, approval gates, execution limits, and safe interruption. Agent adoption is a design choice.
- Close the improvement loop
Review the management system, lessons from incidents, and corrective actions. Pursue external certification when it serves a defined organizational need.
What informed this path
Original guidance informed by these notebook topics, with official resources linked below. The three paths are an editorial guide, not a formal maturity standard.
- 06 - AI Incident Response & Reporting Template_non-c.pdf
- 12 - AI Monitoring, Measurement & Continuous Improvement Template_non-c.pdf
- 13 - AI Internal Audit Template_non-c.pdf
- 14 - AI Management Review Template_non-c.pdf
- 15 - AI Nonconformity & Corrective Action Template_non-c.pdf
- 54_-_Measuring_AI_Security_Effectiveness_Study_Notes.pdf
THE RESOURCE SHELF
Good frameworks. Useful next steps.
Explore the official resources.
AI RMF Playbook
Practical suggestions to help translate AI RMF outcomes into organizational actions.
Put it to work Select actions for your context and assign their owners.
Use alongside the AI RMF; tailor the actions to your system.
FreeOECD AI Principles
Principles for human rights, fairness, transparency, robustness, safety, and accountability.
Put it to work Translate your organization's AI values into policies and decision criteria.
High-level principles need supporting controls and evidence.
FreeISO/IEC 42001
Requirements for establishing, maintaining, and improving an AI management system.
Put it to work Structure policies, responsibilities, operations, reviews, and continual improvement.
Start applying the management approach at any stage; certification is a separate assessment.
Free overview · paid standardGenerative AI Profile
AI RMF companion guidance focused on risks and suggested actions for generative AI.
Put it to work Extend evaluations and risk treatment for your GenAI use cases.
NIST AI 600-1; use with the core AI RMF.
FreeOWASP LLM Top 10
Common LLM application risks including prompt injection, data disclosure, unsafe output handling, and excessive agency.
Put it to work Choose abuse cases and security tests for your application.
A useful risk catalogue; it does not cover every threat.
FreeAI Controls Matrix v1.1
AI control objectives, framework mappings, implementation guidance, and the AI-CAIQ questionnaire.
Put it to work Assign shared control responsibilities and collect supplier or internal evidence.
Select guidance for your role: customer, application, model, platform, or cloud provider.
Free resource · account may be neededMITRE ATLAS
A living knowledge base of adversary tactics, techniques, mitigations, and case studies involving AI.
Put it to work Threat-model your systems and plan adversarial control tests.
Useful earlier too, whenever an AI system needs threat modelling.
FreeEU AI Act information hub
Official policy information and links to the law, guidance, and AI Act Service Desk.
Put it to work Investigate applicability by jurisdiction, system use, and provider/deployer role.
Check current guidance and sector obligations; these stages do not determine legal risk categories.
FreeAISM, AIMS, and AAISM: what is the difference?
- AISM
- Artificial Intelligence Security Management: the discipline of governing and securing AI and managing its risks.
- AIMS
- An Artificial Intelligence Management System: organizational policies, processes, roles, and controls, as structured by ISO/IEC 42001.
- AAISM
- ISACA's Advanced in AI Security Management credential for practitioners. It is distinct from an organization's management system.
How to use this guide
These paths are an original way to navigate resources, informed by my AISM study notebook and checked against official publisher pages. Frameworks can be useful at several stages; the recommendations indicate a starting point, not an eligibility rule.
Adapt actions to each use case, organization, and applicable obligations. Completing a checklist does not establish certification, legal compliance, or system safety. The PDF handouts are original planning templates; source PDFs and paid study materials are not redistributed.
Resource pages reviewed October 3, 2026. Frameworks and regulations evolve; follow the official links for current material.