anchor@anchor-cyber:~$ man ai-rmf-1.0
Anchor Cyber Security LLC
NIST AI RMF 1.0 — Field Reference
Same shape as GOVERN in CSF 2.0, purpose-built for AI systems. For any organization deploying, building on, or assessing AI in production.
$ ai-rmf --glossary
Vocabulary
- AI Actor
- Anyone playing a role in an AI system's lifecycle — designer, developer, deployer, operator, evaluator, end user.
- AI Actor Task
- A concrete lifecycle activity (e.g. "TEVV" — test, evaluation, verification, validation) an AI Actor performs.
- Trustworthy AI
- An AI system judged against seven characteristics (below) rather than a single "safe/unsafe" label.
- Profile
- Same concept as a CSF Profile — which AI RMF outcomes apply, for a specific use case, sector, or system.
- Risk Tolerance
- How much AI risk an organization is willing to accept in pursuit of its objectives — set at the GOVERN level, applied everywhere else.
$ ai-rmf --functions
The Four Functions / 19 categories · 72 subcategories
GOVERN
Cross-cutting, not sequential
Policies, accountability structures, culture, and third-party/supply-chain risk for AI — sits above the other three the same way CSF's GOVERN does.
cat 6sub 19
MAP
Context and categorization
Establish context, categorize the system, understand capabilities/costs/benefits, and characterize impacts to people and society.
cat 5sub 18
MEASURE
Analysis and tracking
Apply methods/metrics, evaluate against the trustworthy characteristics, track risk over time, assess measurement itself.
cat 4sub 22
MANAGE
Prioritize and respond
Act on what MAP and MEASURE found — prioritize risk response, manage third-party AI risk, document treatment and communication plans.
cat 4sub 13
$ ai-rmf --categories -l
Category Map / NIST AI 100-1, Jan 2023
| Code | Category | Subcats |
|---|---|---|
| Govern | ||
| GOVERN 1 | Policies, processes, and practices for mapping, measuring, and managing AI risk are in place and effective | 7 |
| GOVERN 2 | Accountability structures — teams and individuals empowered and trained for AI risk work | 3 |
| GOVERN 3 | Workforce diversity, equity, inclusion, and accessibility in AI risk work | 2 |
| GOVERN 4 | Organizational culture that considers and communicates AI risk | 3 |
| GOVERN 5 | Robust engagement with relevant AI Actors | 2 |
| GOVERN 6 | Policies for third-party software, data, and supply chain AI risk | 2 |
| Map | ||
| MAP 1 | Context is established and understood | 6 |
| MAP 2 | Categorization of the AI system is performed | 3 |
| MAP 3 | AI capabilities, usage, goals, and expected costs/benefits are understood | 5 |
| MAP 4 | Risks and benefits mapped for all components, including third-party software and data | 2 |
| MAP 5 | Impacts to individuals, groups, communities, and society are characterized | 2 |
| Measure | ||
| MEASURE 1 | Appropriate methods and metrics identified and applied | 3 |
| MEASURE 2 | AI systems evaluated for the seven trustworthy characteristics | 13 |
| MEASURE 3 | Mechanisms for tracking identified risks over time are in place | 3 |
| MEASURE 4 | Feedback on measurement efficacy is gathered and assessed | 3 |
| Manage | ||
| MANAGE 1 | Risk from MAP and MEASURE is prioritized, responded to, and managed | 4 |
| MANAGE 2 | Strategies to maximize benefit and minimize harm are planned and documented | 4 |
| MANAGE 3 | Third-party AI risks and benefits are managed | 2 |
| MANAGE 4 | Risk treatment, response, recovery, and communication plans are documented and monitored | 3 |
$ ai-rmf --trustworthy
Seven Trustworthy Characteristics / what MEASURE actually evaluates
Valid & ReliableAccurate for its intended purpose, consistent under expected conditions.
SafeDoesn't endanger human life, health, property, or the environment.
Secure & ResilientWithstands adversarial manipulation and recovers from disruption.
Accountable & TransparentClear ownership of outcomes; appropriate disclosure about the system.
Explainable & InterpretableA human can understand the output and why the system produced it.
Privacy-EnhancedAnonymity, confidentiality, and user control over data are safeguarded.
Fair, Harmful Bias ManagedEquity considered across systemic, computational, and human-cognitive bias.
$ ai-rmf --apply
Where AI RMF Shows Up in Practice
- AI-inspection or filtering tools.A tool that inspects, filters, or makes decisions about AI traffic is itself an AI Actor — MAP 1/2 apply to it before they apply to a customer's downstream AI usage.
- Client AI adoption reviews.Use GOVERN 1/2 to check whether a client has any AI risk policy or accountable owner before they roll out an LLM tool company-wide.
- Your own AI use policy.Cross-check it against GOVERN 4 (culture) and GOVERN 6 (third-party AI risk) — most internal AI policies only cover acceptable use, not vendor risk.
- Incident review involving an AI system.MEASURE 3 and MANAGE 4 are the two categories that ask "are we tracking this risk over time," not just "did we fix it once."