{
  "name": "Enterprise AI Risk Matrix",
  "description": "A framework for classifying and prioritizing enterprise AI and agentic-AI risks by likelihood and impact, with governance controls.",
  "url": "https://gianlucabusato.com/datasets/enterprise-ai-risk-matrix/",
  "creator": {
    "name": "Gianluca Busato",
    "id": "https://gianlucabusato.com/#person"
  },
  "dateModified": "2026-07-15",
  "inLanguage": "en",
  "columns": [
    "Risk category",
    "Example",
    "Typical impact",
    "Primary control"
  ],
  "records": [
    {
      "Risk category": "Unauthorized action",
      "Example": "Agent executes an action outside its mandate",
      "Typical impact": "High",
      "Primary control": "Scoped permissions + policy enforcement"
    },
    {
      "Risk category": "Identity spoofing",
      "Example": "Agent acts under a false or borrowed identity",
      "Typical impact": "High",
      "Primary control": "Verifiable agent identity"
    },
    {
      "Risk category": "Prompt injection",
      "Example": "Malicious input overrides instructions",
      "Typical impact": "High",
      "Primary control": "Input mediation + guard agents"
    },
    {
      "Risk category": "Data leakage",
      "Example": "Sensitive data exposed via output or tool call",
      "Typical impact": "High",
      "Primary control": "Data governance + output filtering"
    },
    {
      "Risk category": "Hallucinated decision",
      "Example": "Action based on fabricated information",
      "Typical impact": "Medium",
      "Primary control": "Grounding + human-in-the-loop"
    },
    {
      "Risk category": "Cascading failure",
      "Example": "One agent's error propagates across systems",
      "Typical impact": "High",
      "Primary control": "Containment + blast-radius limits"
    },
    {
      "Risk category": "Non-compliance",
      "Example": "Action violates regulation or policy",
      "Typical impact": "High",
      "Primary control": "Compliance-by-design + audit trail"
    },
    {
      "Risk category": "Opacity",
      "Example": "No record of why an action occurred",
      "Typical impact": "Medium",
      "Primary control": "Operational transparency + logging"
    }
  ]
}