AI Safety

Describe your use case with the six sliders

Each slider runs from 1 (lowest risk) to 10 (highest risk). As you move one, the text below it describes exactly what that level means and gives a concrete example. Pick the level whose description best matches what you are actually planning — not what you hope to achieve later.

• Scope — how much of the task the AI performs, from a single suggestion to fully autonomous operation.
• Human Oversight — whether and when a person reviews or approves the AI's work.
• Transparency — how visible and auditable the AI's actions and reasoning are.
• Stakes — how much harm an error causes, and how reversible that harm is.
• Bias & Fairness — the risk of discriminatory or inequitable outcomes across groups.
• Privacy — how sensitive the personal data involved is, and what protections are in place.

Read the risk score and the shape of the chart

The score is the average of all six dimensions, banded as Minimal, Low, Moderate, High, or Critical. Pay attention to the shape of the chart, not just the number: a single dimension at 9 is a real problem even when the average looks comfortable. A spike on one axis tells you where to focus mitigation.

Compare against the closest real-world analog

The tool matches your profile against a library of well-understood AI use cases and shows the nearest one, with a similarity percentage. This is a sanity check. If your proposed internal tool lands next to an autonomous fraud-blocking system, that mismatch is worth a conversation before you build anything.

Work through the NIST AI RMF guidance

The panel below the tool maps your profile to the four functions of the NIST AI Risk Management Framework, and its recommendations scale with your score:

• Govern — the policy, ownership, and accountability you need in place.
• Map — the context to document and the stakeholders to identify.
• Measure — the testing, bias audits, and monitoring to perform.
• Manage — the mitigations, escalation paths, and incident response to build.

High scores on individual dimensions add targeted recommendations — elevated privacy risk triggers a Privacy Impact Assessment, elevated bias risk calls for a fairness audit, and so on.

Use it to structure the conversation

The most useful move is to score a use case twice: once as proposed, and once with mitigations applied. Watching the score drop when you add human review or an audit log makes the value of a control concrete, and turns an abstract debate into a specific list of things to change.

What it measures

Six sliders, each scored from 1 to 10, with ten written levels and a concrete example at every level so that scoring is a matter of recognition rather than guesswork.

What it produces

• A banded risk score — Minimal, Low, Moderate, High, or Critical.
• A six-axis chart that shows the shape of the risk, not just its magnitude.
• The closest real-world analog — the nearest match from a library of twenty well-understood AI systems, with a similarity percentage.
• NIST AI RMF guidance across all four functions — Govern, Map, Measure, and Manage — scaled to the score, with additional recommendations triggered when an individual dimension runs high.

Benefits