AI Guardrails · Schema

Guardrail Violation

A single event emitted by a guardrail when input, output, retrieval, dialog, or execution content matched a rule. Designed to be portable across Guardrails AI, NVIDIA NeMo Guardrails, Lakera, Azure Prompt Shields, AWS Bedrock Guardrails, OpenAI Moderation, and Google Model Armor.

AI SafetyAI SecurityContent ModerationGuardrailsJailbreak DetectionLLM SecurityPII DetectionPrompt InjectionResponsible AI

Properties

Name	Type	Description
id	string	Globally unique identifier for the violation event.
policyId	string	Identifier of the policy that fired.
policyVersion	string	Version of the policy at the time of evaluation.
ruleId	string	Identifier of the specific rule that matched.
vendor	string	Guardrail vendor that produced the event.
direction	string
category	string
severity	string
action	string	Outcome applied as a result of this match.
score	number	Detector confidence score where applicable.
timestamp	string
request	object	Metadata about the originating request.
content	object	The flagged content. Implementations should hash or redact rather than store raw PII.
detector	object
remediation	object	What the application should surface in response to the violation.
tags	array

View JSON Schema on GitHub

JSON Schema

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://raw.githubusercontent.com/api-evangelist/guardrails/main/json-schema/guardrail-violation-schema.json",
  "title": "Guardrail Violation",
  "description": "A single event emitted by a guardrail when input, output, retrieval, dialog, or execution content matched a rule. Designed to be portable across Guardrails AI, NVIDIA NeMo Guardrails, Lakera, Azure Prompt Shields, AWS Bedrock Guardrails, OpenAI Moderation, and Google Model Armor.",
  "type": "object",
  "required": ["id", "policyId", "ruleId", "direction", "category", "action", "timestamp"],
  "properties": {
    "id": {
      "type": "string",
      "description": "Globally unique identifier for the violation event.",
      "examples": ["urn:guardrail-violation:01HG3F2N6PMA7Q5"]
    },
    "policyId": {
      "type": "string",
      "description": "Identifier of the policy that fired."
    },
    "policyVersion": {
      "type": "string",
      "description": "Version of the policy at the time of evaluation."
    },
    "ruleId": {
      "type": "string",
      "description": "Identifier of the specific rule that matched."
    },
    "vendor": {
      "type": "string",
      "description": "Guardrail vendor that produced the event."
    },
    "direction": {
      "type": "string",
      "enum": ["input", "output", "retrieval", "dialog", "execution"]
    },
    "category": {
      "type": "string",
      "enum": [
        "prompt-injection",
        "jailbreak",
        "indirect-prompt-injection",
        "pii",
        "sensitive-information",
        "content-safety",
        "hate",
        "harassment",
        "self-harm",
        "sexual",
        "violence",
        "hallucination",
        "contextual-grounding",
        "denied-topic",
        "competitor-mention",
        "profanity",
        "toxic-language",
        "malicious-url",
        "data-exfiltration",
        "structured-output",
        "tool-misuse",
        "agent-goal-hijack",
        "policy-violation"
      ]
    },
    "severity": {
      "type": "string",
      "enum": ["info", "low", "medium", "high", "critical"]
    },
    "action": {
      "type": "string",
      "enum": ["allowed", "blocked", "redacted", "transformed", "logged", "queued-for-review"],
      "description": "Outcome applied as a result of this match."
    },
    "score": {
      "type": "number",
      "minimum": 0,
      "maximum": 1,
      "description": "Detector confidence score where applicable."
    },
    "timestamp": {
      "type": "string",
      "format": "date-time"
    },
    "request": {
      "type": "object",
      "description": "Metadata about the originating request.",
      "properties": {
        "id": {"type": "string"},
        "application": {"type": "string"},
        "model": {"type": "string"},
        "userId": {"type": "string"},
        "sessionId": {"type": "string"},
        "tenantId": {"type": "string"}
      }
    },
    "content": {
      "type": "object",
      "description": "The flagged content. Implementations should hash or redact rather than store raw PII.",
      "properties": {
        "sample": {
          "type": "string",
          "description": "Representative excerpt of the flagged content; should be redacted/hashed for sensitive categories."
        },
        "language": {"type": "string"},
        "tokenCount": {"type": "integer", "minimum": 0},
        "spans": {
          "type": "array",
          "description": "Character offsets of matched substrings within the original content.",
          "items": {
            "type": "object",
            "required": ["start", "end"],
            "properties": {
              "start": {"type": "integer", "minimum": 0},
              "end": {"type": "integer", "minimum": 0},
              "label": {"type": "string"},
              "replacement": {"type": "string"}
            }
          }
        }
      }
    },
    "detector": {
      "type": "object",
      "properties": {
        "type": {"type": "string"},
        "model": {"type": "string"},
        "version": {"type": "string"},
        "latencyMs": {"type": "integer", "minimum": 0}
      }
    },
    "remediation": {
      "type": "object",
      "description": "What the application should surface in response to the violation.",
      "properties": {
        "userMessage": {"type": "string"},
        "internalNote": {"type": "string"},
        "escalationChannel": {"type": "string"}
      }
    },
    "tags": {
      "type": "array",
      "items": {"type": "string"}
    }
  }
}