{
  "node_id": "nistir-8312-explainable-ai-principles",
  "title": "NISTIR 8312 Four Principles of Explainable Artificial Intelligence",
  "domain": "AI Governance & Law",
  "version": "1.0.0",
  "last_updated": "2021-09-01",
  "bluf": "This document introduces four principles for explainable artificial intelligence (AI) that comprise fundamental properties for explainable AI systems. For AI systems that are intended or required to be explainable, it is proposed that they adhere to these principles. First, a system must deliver accompanying evidence or reasons for its outcomes and processes (Explanation). Second, these explanations must be understandable to the individual users they are intended for (Meaningful). Third, the explanation must correctly reflect the system’s actual process for generating the output (Explanation Accuracy). Finally, the system must only operate under the conditions for which it was designed and when it reaches sufficient confidence in its output (Knowledge Limits).\n\nThese principles were developed to encompass the multidisciplinary nature of explainable AI and are heavily influenced by the AI system’s interaction with the human recipient. The requirements of a given situation, the task at hand, and the consumer will all influence the type of explanation deemed appropriate. These situations can include regulatory and legal requirements, quality control, and customer relations. The principles allow for defining the contextual factors to consider for an explanation and act as a roadmap for future measurement and evaluation activities. This work is part of a larger NIST portfolio around trustworthy AI, which also includes characteristics like accuracy, privacy, reliability, robustness, safety, security, mitigation of harmful bias, transparency, fairness, and accountability.",
  "paywall": {
    "status": "LOCKED",
    "unlock_cost_usd": "0.01",
    "skyfire_id": "41779894-ece2-4163-9761-b3b1b76e19b0"
  },
  "crosswalks": {
    "_available_keys": [
      "nist_framework",
      "iso_standard",
      "ai_overlay_2026",
      "industry_mapping"
    ],
    "_note": "Full crosswalk values included in vault response"
  },
  "dependencies": [
    "nist-ai-rmf-1-0",
    "oecd-ai-principles",
    "nist-sp-1270-managing-ai-bias"
  ],
  "primary_citations_count": 7
}