Examine possible biases in datasets

AIR-0370617F00 · version 2 · Fairness · candidate interpretation

High-risk AI providers must examine possible biases in datasets.

Situation & purpose

An AI system or employee activity involves possible biases in datasets. Confirm the actor and source conditions before using this template.

Provide a traceable control for the action: examine possible biases in datasets.

Examples

Meets the rule: High-risk AI providers performs the required action "examine" for possible biases in datasets within these source conditions: Training, validation and testing data, and retains evidence.
Does not meet the rule: High-risk AI providers omits the required action "examine" for possible biases in datasets while those source conditions apply.
Outside scope: The actor, activity, jurisdiction, or applicability conditions in this source variant do not match.

Evidence & enforcement

Dated record of the action
Responsible actor and approval, where required
Applicable source edition and scope

Human or authorized system adapter required. Machine-executable: no. Convert to a reviewed organization contract and validate the adapter before deployment.

Limits

A rule card is an interpretation aid, not an applicability determination.

Equivalent intent does not make different source obligations interchangeable.

Owner review

Record your interpretation decision. This does not adopt a policy for an organization.

Review reason

Owner sign-in

JSON rule card

{
  "id": "AIR-0370617F00",
  "version": 2,
  "title": "Examine possible biases in datasets",
  "text": "High-risk AI providers must examine possible biases in datasets.",
  "topic": "Fairness",
  "intent": "examine possible biases in datasets",
  "status": "candidate",
  "semanticStatus": "human-validation-required",
  "variants": [
    {
      "sourceId": "eu-ai-act",
      "control": "Article 10(2)(f)",
      "sourceVersion": "Regulation 2024/1689; 2026 amendments tracked separately",
      "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng",
      "actor": "High-risk AI providers",
      "modality": "must",
      "action": "examine",
      "object": "possible biases in datasets",
      "condition": "Training, validation and testing data",
      "exception": "Applicable statutory provisions",
      "timing": null,
      "authority": "binding-within-scope",
      "jurisdiction": "European Union",
      "checkedAt": "2026-10-03",
      "status": "interpretation-needs-review",
      "examples": {
        "meetsRule": "High-risk AI providers performs the required action \"examine\" for possible biases in datasets within these source conditions: Training, validation and testing data, and retains evidence.",
        "doesNotMeetRule": "High-risk AI providers omits the required action \"examine\" for possible biases in datasets while those source conditions apply.",
        "outOfScope": "The actor, activity, jurisdiction, or applicability conditions in this source variant do not match."
      }
    }
  ],
  "situation": "An AI system or employee activity involves possible biases in datasets. Confirm the actor and source conditions before using this template.",
  "purpose": "Provide a traceable control for the action: examine possible biases in datasets.",
  "examples": {
    "meetsRule": "High-risk AI providers performs the required action \"examine\" for possible biases in datasets within these source conditions: Training, validation and testing data, and retains evidence.",
    "doesNotMeetRule": "High-risk AI providers omits the required action \"examine\" for possible biases in datasets while those source conditions apply.",
    "outOfScope": "The actor, activity, jurisdiction, or applicability conditions in this source variant do not match."
  },
  "evidence": [
    "Dated record of the action",
    "Responsible actor and approval, where required",
    "Applicable source edition and scope"
  ],
  "enforcement": {
    "mode": "policy-template",
    "adapter": "Human or authorized system adapter required",
    "machineExecutable": false
  },
  "limitations": [
    "A rule card is an interpretation aid, not an applicability determination.",
    "Equivalent intent does not make different source obligations interchangeable."
  ],
  "createdAt": "2026-10-03",
  "updatedAt": "2026-10-03"
}

Back to registry