RiskFramework.org
RiskFramework.org
GitHub LinkedIn

Practical tools for better risk decisions.

Tools

MAP v1.3 Stable CC BY-SA 4.0 Updated 2026
A folded map with a pin marked AI

AI Risk Map

A thought-starter for identifying AI-specific risk factors.

What it is

A single-page visual reference that helps teams widen their perspective, surface blind spots, and find overlooked AI risks.

What it covers

The risks that come with your organization's own use of AI — what you take on when you build, buy, and deploy AI.

196 risk factors across 55 categories and 12 domains, organized by who owns the risk in the enterprise, not by technology: governance, strategy, data, model development, security, safety, operations, legal, third-party, workforce, financial, and autonomous action.

Column 3 of 12, one factor circled · full poster 48 × 36 in Full-size SVG ↗ PDF ↗

Get it

  • PosterPDF · PNG · SVG
  • Taxonomy workbookXLSX
  • Flat tableCSV
  • Corpus for AI assistantsMD
Get the files on GitHub

Updated in v1.3: eight new factors and four narrowed ones. See what changed before updating an existing mapping.

Every column on the poster follows the same three tiers. Here’s how to read one.

How to read the AI Risk Map Three tiers: a numbered, colored domain block; a category label with a rule in the domain color; risk factors hanging from a line, one carrying a cross-reference pill to domain 8. HOW TO READ THE MAP 01 Domain the enterprise function that owns the risk Category a group of related factors Risk factor: a condition, failure mode, or event that contributes to risk One side of a split pair »8 = other side in domain 8, in its color Color is the domain. Position and size say nothing about likelihood or severity. How to read the AI Risk Map Three tiers: a numbered, colored domain block; a category label with a rule in the domain color; risk factors hanging from a line, one carrying a cross-reference pill to domain 8. HOW TO READ THE MAP 01 Domain The enterprise function that owns the risk Category A group of related factors Risk factor: a condition, failure mode, or event that contributes to risk One side of a split pair »8 = other side in domain 8, in its color Color is the domain. Position and size say nothing about likelihood or severity.

Why I made it

Years ago, when organizations were beginning to adopt public cloud, my boss brought a cloud risk poster into the office. It wasn't an assessment framework or an exhaustive inventory. It was a visual reminder to widen our perspective — and it sparked better risk conversations.

When generative AI became part of our everyday conversations, I found myself wishing I had something similar for AI. The AI Risk Map is my attempt to create it.

How to use it

  1. In a workshopPut the poster up and ask which factors nobody has discussed. The PDF prints at 48 × 36 in, a standard large-format size at any print shop.
  2. With an AI assistantLoad the markdown file and ask domain-specific questions.
  3. In your own toolingMap against the CSV using the stable IDs AIRM-001 to AIRM-197.

What counts as a “risk factor” here

The entries mix conditions, failure modes, threat events, and loss events. They are not well-formed loss-event statements in the FAIR sense — scope one into a proper risk scenario before using it in analysis. Some factors scope cleanly into CRQ/FAIR scenarios; others are board-narrative.

What it is not

  • Not a control framework. It names risk factors, never controls.
  • Not a threat register. Other people's malicious use of AI against you, such as deepfakes and AI-enabled fraud, is deliberately out of scope. How competitors' and customers' AI changes the value of your own strategy is in scope.
  • Not a scoring tool. Position, order and ID number say nothing about likelihood or severity.

Limitations

  • A point-in-time view of a fast-moving field.
  • The map helps surface possibilities; it doesn't determine which risks matter most to your organization.
  • The twelve domains assume a typical enterprise structure and may not match yours.
  • It seeds threat modeling and risk assessment; it complements, it does not replace, system-specific models.
Socrates, hand on chin
I had to make a lot of hard inclusion decisions.

How it was built

The map lists the risks AI introduces, or amplifies in a specific and nameable way. Generic IT, cyber, and vendor risk that exists identically without AI stays off — that one constraint is what keeps the map focused. A candidate earned a place by clearing any one of three gates:

AI-specific
Remove AI and the risk vanishes.
Management delta
The risk is not new, but the way it must be managed is — and the delta has to be specific.
Corroboration
Two or more independent AI standards name it.

Sources span regulation and supervisory guidance, standards (NIST, ISO/IEC), threat frameworks and academic taxonomies. All 196 factors carry citations verified against primary texts.

A few entries may look like duplicates. They aren't. Some are the same loss seen from two owners' columns, such as a hallucination and the harm it causes. These split pairs are marked »n on the poster and counted once. Others share an outcome but happen in different ways, such as data extracted by an attacker and data disclosed by accident. Those are separate risks, linked in the workbook, and counted separately.

The taxonomy workbook is the audit trail: every factor's sources, a scope-decisions sheet recording what was excluded and why, and a change log across versions. If you're wondering why something isn't on the map, look there first.