The NIST AI RMF 1.0
Attaches to every line through the MEASURE items on performance under deployment conditions, to named tools through third-party monitoring, and to the lines that decide through the oversight and determination items.
On the register, tick "NIST AI RMF" and these rows appear on every line the regime reaches. Source framework: NIST AI RMF 1.0.
Which lines it reaches, and with what
| On every classified line | MS-2.3, MS-2.5. MEASURE: performance demonstrated under conditions like the deployment, and validity and reliability shown before deployment with the limits written down. |
|---|---|
| On lines that name a tool | MN-3.1. MANAGE: the third-party tool is monitored on an ongoing basis, not assessed once at procurement. |
| On lines that decide (with review or alone) | MP-3.5, MN-1.1. MAP and MANAGE: the human oversight process defined and assessed for whether it is exercisable at the pace of the line, and the recorded determination that the system achieves its purpose. |
The record it expects
| informs | The tool produces a summary, a ranking or a flag that a person reads before deciding; the person owns the decision. The record holds: who read the output and what it informed; date and the tool version; the output kept with the decision it informed. |
|---|---|
| drafts | The tool produces the draft of a document, a figure or a position that a person edits and signs; the person owns the content. The record holds: who edited and who signed; date and the tool version; the draft kept beside the signed version; what was changed, classed as omission, misstatement or wording. |
| decides with review | The tool takes the decision and a named person reviews every case or a stated share before the effect; the reviewer can change the outcome. The record holds: reviewer, role and the competence that qualifies them (Art. 26); date and the tool version; sample reviewed: the count out of the total, and the share; overrides: count and rate; escalations and to whom; log reference and retention period (Art. 12, Art. 26); input data checked as relevant and representative (Art. 26). |
| decides alone | The tool takes the decision and the effect follows with no person in between; a person may audit afterwards. The record holds: who audits the line afterwards, role and competence (Art. 26); date, the tool version and the volume decided; audit sample: the count out of the total, and the share; reversals after audit: count and rate; escalations and to whom; log reference and retention period (Art. 12, Art. 26); input data checked as relevant and representative (Art. 26); incidents reported to the provider and the date (Art. 26). |
The clauses, quoted
5 of 72 in the frameworkRequirement text in our own words, read clause by clause from the published standards: our statement of each clause, not the instrument verbatim.
AI RMF MS-2.3 AI system performance or assurance criteria are measured qualitatively or quantitatively and demonstrated for conditions similar to deployment settings, and measures are documentedAI system performance or assurance criteria are measured qualitatively or quantitatively and demonstrated for conditions similar to deployment setting(s). Measures are documented. Performance is demonstrated under conditions resembling deployment rather than only on held-out test data, and the measures are recorded.
Common gap: Performance demonstrated only in silico on a static test set
Source framework: NIST AI RMF 1.0
AI RMF MS-2.5 The AI system to be deployed is demonstrated to be valid and reliable, and limitations of the generalizability beyond the conditions under which the technology was developed are documentedThe AI system to be deployed is demonstrated to be valid and reliable. Limitations of the generalizability beyond the conditions under which the technology was developed are documented. Validity and reliability are demonstrated before deployment, and the boundary beyond which the demonstration does not carry is written down.
Common gap: Validity claimed from training performance rather than independent validation
Source framework: NIST AI RMF 1.0
AI RMF MN-3.1 AI risks and benefits from third-party resources are regularly monitored, and risk controls are applied and documentedAI risks and benefits from third-party resources are regularly monitored, and risk controls are applied and documented. Third-party data, model, software and hardware dependencies are monitored on an ongoing basis, not assessed once at procurement, and the controls applied are recorded.
Common gap: Third parties assessed at onboarding and not monitored afterwards
Source framework: NIST AI RMF 1.0
AI RMF MP-3.5 Processes for human oversight are defined, assessed, and documented in accordance with organizational policies from the GOVERN functionProcesses for human oversight are defined, assessed, and documented in accordance with organizational policies from GOVERN function. Oversight is designed for the specific configuration, assessed for whether it is exercisable in practice, and consistent with the governing policy rather than invented per project.
Common gap: Oversight designed at a volume the reviewer cannot sustain
Source framework: NIST AI RMF 1.0
AI RMF MN-1.1 A determination is made as to whether the AI system achieves its intended purpose and stated objectives and whether its development or deployment should proceedA determination is made as to whether the AI system achieves its intended purpose and stated objectives and whether its development or deployment should proceed. A recorded determination weighs measured risk against benefit and decides whether to proceed, so that not proceeding remains an available and evidenced outcome.
Common gap: Determination recorded as approval with no evidence weighed
Source framework: NIST AI RMF 1.0
See what it attaches to your list
Paste the list, tick the regime, and every line it reaches carries these rows. Eight lines free, no account.
Build my decision register