AI Guardrails

aid: guardrails
name: AI Guardrails
description: >-
  AI Guardrails are runtime and design-time controls that screen the inputs and
  outputs of large language model (LLM) applications and AI agents. They detect
  and block prompt injection, jailbreak attempts, personally identifiable information
  (PII) leakage, toxic or unsafe content, hallucinations, and policy violations,
  and they validate structured outputs against schemas. This topic repository
  catalogs the vendor and open-source landscape — provider-native guardrails
  (AWS Bedrock, Azure Content Safety, Google Model Armor, OpenAI Moderation),
  third-party AI security platforms (Lakera, HiddenLayer, Cisco AI Defense,
  Lasso Security, PromptArmor, Wallarm), and open-source frameworks (Guardrails
  AI, NVIDIA NeMo Guardrails, DeepEval) — and provides a shared vocabulary,
  JSON Schema for policy and violation records, JSON-LD context, and example
  payloads.
type: Index
image: https://kinlane-productions.s3.amazonaws.com/apis-json/apis-json-logo.jpg
tags:
  - AI Safety
  - AI Security
  - Content Moderation
  - Guardrails
  - Jailbreak Detection
  - LLM Security
  - PII Detection
  - Prompt Injection
  - Responsible AI
url: https://raw.githubusercontent.com/api-evangelist/guardrails/refs/heads/main/apis.yml
humanURL: https://github.com/api-evangelist/guardrails
created: '2026-05-22'
modified: '2026-05-22'
specificationVersion: '0.19'
apis:
  - aid: guardrails:guardrails-ai
    name: Guardrails AI
    description: >-
      Open-source Python framework and commercial Hub for adding programmable
      input/output validators to LLM applications. Validators cover regex, PII,
      toxic language, competitor mentions, jailbreak attempts, and structured-output
      schema enforcement. The commercial side ships Snowglobe for synthetic data
      and an AI Reliability Platform.
    humanURL: https://www.guardrailsai.com/
    baseURL: https://api.guardrailsai.com/
    tags:
      - Hallucination
      - Jailbreak Detection
      - Open Source
      - PII Detection
      - Structured Output
      - Validator Hub
    properties:
      - type: Homepage
        url: https://www.guardrailsai.com/
      - type: Documentation
        url: https://www.guardrailsai.com/docs
      - type: GitHub
        url: https://github.com/guardrails-ai/guardrails
      - type: Hub
        url: https://hub.guardrailsai.com/
      - type: License
        url: https://github.com/guardrails-ai/guardrails/blob/main/LICENSE
      - type: PyPI
        url: https://pypi.org/project/guardrails-ai/
      - type: x-deployment
        value: SDK
      - type: x-threat-categories
        value: prompt-injection,pii,jailbreak,hallucination,structured-output
  - aid: guardrails:nvidia-nemo-guardrails
    name: NVIDIA NeMo Guardrails
    description: >-
      Open-source toolkit for adding programmable guardrails to LLM-based
      conversational systems. Implements five rail types — input, output, dialog,
      retrieval, and execution — using the Colang modeling language. Integrates
      with OpenAI, LLaMa, Falcon, Vicuna, and Mosaic models.
    humanURL: https://docs.nvidia.com/nemo/guardrails/
    tags:
      - Colang
      - Dialog Rails
      - Input Rails
      - NVIDIA
      - Open Source
      - Output Rails
      - Retrieval Rails
    properties:
      - type: Homepage
        url: https://developer.nvidia.com/nemo-guardrails
      - type: Documentation
        url: https://docs.nvidia.com/nemo/guardrails/
      - type: GitHub
        url: https://github.com/NVIDIA/NeMo-Guardrails
      - type: PyPI
        url: https://pypi.org/project/nemoguardrails/
      - type: License
        url: https://github.com/NVIDIA/NeMo-Guardrails/blob/main/LICENSE
      - type: x-deployment
        value: SDK
      - type: x-threat-categories
        value: prompt-injection,jailbreak,content-safety,hallucination,dialog-policy
  - aid: guardrails:lakera-ai
    name: Lakera AI
    description: >-
      AI-native security platform protecting generative AI applications and agents
      from prompt injection, data leakage, toxic content, and compliance risks.
      Products include Lakera Guard (runtime API), Workforce AI Security, AI Agent
      Security, AI Red Teaming, and Gandalf (red-teaming training game).
    humanURL: https://www.lakera.ai/
    baseURL: https://api.lakera.ai/
    tags:
      - AI Red Teaming
      - Data Leakage
      - Multilingual
      - Prompt Injection
      - Runtime API
      - Third-Party Vendor
    properties:
      - type: Homepage
        url: https://www.lakera.ai/
      - type: Documentation
        url: https://docs.lakera.ai/docs/quickstart
      - type: APIReference
        url: https://docs.lakera.ai/docs/api
      - type: Changelog
        url: https://platform.lakera.ai/docs/changelog
      - type: Customers
        url: https://www.lakera.ai/customers
      - type: x-deployment
        value: API
      - type: x-threat-categories
        value: prompt-injection,pii,content-safety,jailbreak,data-leakage
      - type: x-scale
        value: 1M+ secured transactions/app/day; sub-50ms latency; 100+ languages
  - aid: guardrails:microsoft-azure-prompt-shields
    name: Microsoft Azure AI Content Safety — Prompt Shields
    description: >-
      Unified API in Azure AI Content Safety that detects and blocks adversarial
      user-prompt attacks and indirect document attacks on LLMs. Replaces the
      earlier Jailbreak risk detection service. Detects role-play, system-rule
      changes, conversation-mockup attacks, encoding attacks, and document-borne
      indirect prompt injection.
    humanURL: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/jailbreak-detection
    tags:
      - Azure
      - Content Safety
      - Document Attacks
      - Indirect Prompt Injection
      - Jailbreak Detection
      - Microsoft
      - Provider-Native
    properties:
      - type: Documentation
        url: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/jailbreak-detection
      - type: Overview
        url: https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview
      - type: APIReference
        url: https://learn.microsoft.com/en-us/rest/api/contentsafety/
      - type: Pricing
        url: https://azure.microsoft.com/en-us/pricing/details/cognitive-services/content-safety/
      - type: x-deployment
        value: Cloud Service
      - type: x-threat-categories
        value: prompt-injection,jailbreak,indirect-prompt-injection,content-safety
  - aid: guardrails:aws-bedrock-guardrails
    name: Amazon Bedrock Guardrails
    description: >-
      Configurable safeguards on Amazon Bedrock providing content filters,
      denied topics, word filters, sensitive-information (PII) filters,
      contextual grounding checks (hallucination detection in RAG), and
      Automated Reasoning checks. Invokable inline with foundation models or
      standalone via the ApplyGuardrail API.
    humanURL: https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html
    tags:
      - AWS
      - Bedrock
      - Contextual Grounding
      - Denied Topics
      - PII Detection
      - Provider-Native
      - RAG
    properties:
      - type: Documentation
        url: https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html
      - type: APIReference
        url: https://docs.aws.amazon.com/bedrock/latest/APIReference/API_Operations_Amazon_Bedrock.html
      - type: Pricing
        url: https://aws.amazon.com/bedrock/pricing/
      - type: x-deployment
        value: Cloud Service
      - type: x-threat-categories
        value: content-safety,denied-topics,pii,prompt-injection,hallucination
  - aid: guardrails:openai-moderation
    name: OpenAI Moderation API
    description: >-
      Free OpenAI endpoint that classifies text and images across harmful-content
      categories including sexual, hate, harassment, self-harm, violence, and
      illicit. Multimodal moderation model is omni-moderation-latest.
    humanURL: https://platform.openai.com/docs/guides/moderation
    baseURL: https://api.openai.com/v1/moderations
    tags:
      - Content Moderation
      - Free Tier
      - Multimodal
      - OpenAI
      - Provider-Native
    properties:
      - type: Documentation
        url: https://platform.openai.com/docs/guides/moderation
      - type: APIReference
        url: https://platform.openai.com/docs/api-reference/moderations
      - type: Pricing
        url: https://openai.com/api/pricing/
      - type: x-deployment
        value: API
      - type: x-threat-categories
        value: content-safety,hate,harassment,self-harm,sexual,violence
  - aid: guardrails:google-model-armor
    name: Google Cloud Model Armor
    description: >-
      Google Cloud service that screens LLM prompts and responses for prompt
      injection, jailbreak attacks, sensitive data (PII, credit cards, SSNs,
      API keys), harmful content (hate, harassment, sexual, dangerous, CSAM),
      and malicious URLs. Stateless inspector with Inspect-Only and Inspect-and-Block
      enforcement modes. Integrates with Apigee, Gemini Enterprise, GKE, Vertex
      Agent Platform, and LangChain.
    humanURL: https://docs.cloud.google.com/security-command-center/docs/model-armor-overview
    tags:
      - Apigee
      - Google Cloud
      - Malicious URLs
      - PII Detection
      - Prompt Injection
      - Provider-Native
      - Vertex AI
    properties:
      - type: Documentation
        url: https://docs.cloud.google.com/security-command-center/docs/model-armor-overview
      - type: SecurityCommandCenter
        url: https://cloud.google.com/security-command-center
      - type: x-deployment
        value: Cloud Service
      - type: x-threat-categories
        value: prompt-injection,jailbreak,pii,content-safety,malicious-urls
  - aid: guardrails:hiddenlayer
    name: HiddenLayer
    description: >-
      AI security platform spanning AI Discovery, AI Supply Chain Security, AI
      Attack Simulation, and AI Runtime Security. Defends against prompt injection,
      jailbreaks, model manipulation, data leakage, and supply-chain compromise
      using deterministic classifiers that operate outside the LLM inference path.
    humanURL: https://hiddenlayer.com/
    tags:
      - Adversarial ML
      - AI Detection and Response
      - Model Scanner
      - Runtime Security
      - Supply Chain
      - Third-Party Vendor
    properties:
      - type: Homepage
        url: https://hiddenlayer.com/
      - type: Platform
        url: https://hiddenlayer.com/aisec-platform/
      - type: x-deployment
        value: Platform
      - type: x-threat-categories
        value: prompt-injection,jailbreak,model-manipulation,data-leakage,supply-chain
  - aid: guardrails:cisco-ai-defense
    name: Cisco AI Defense (formerly Robust Intelligence)
    description: >-
      Cisco AI Defense — the productization of the Robust Intelligence acquisition —
      provides AI model validation, runtime protection, and continuous algorithmic
      red teaming for production LLM and ML applications. Covers prompt injection,
      jailbreak, model vulnerability scanning, and policy enforcement at the network
      and application layer.
    humanURL: https://www.cisco.com/site/us/en/products/security/ai-defense/index.html
    tags:
      - Algorithmic Red Teaming
      - Cisco
      - Model Validation
      - Network-Layer
      - Robust Intelligence
      - Third-Party Vendor
    properties:
      - type: Homepage
        url: https://www.cisco.com/site/us/en/products/security/ai-defense/index.html
      - type: x-deployment
        value: Platform
      - type: x-threat-categories
        value: prompt-injection,jailbreak,model-vulnerability,policy-enforcement
  - aid: guardrails:lasso-security
    name: Lasso Security
    description: >-
      Enterprise AI security platform providing Intent Security, AI Discovery
      and AI-BOM, Automated AI Red Teaming, AI Detection and Response, and
      Runtime Enforcement. Deploys at the proxy, API, or AI gateway layer.
    humanURL: https://www.lasso.security/
    tags:
      - AI-BOM
      - AI Discovery
      - AI Gateway
      - Enterprise
      - Runtime Enforcement
      - Third-Party Vendor
    properties:
      - type: Homepage
        url: https://www.lasso.security/
      - type: Customers
        url: https://www.lasso.security/customers
      - type: x-deployment
        value: Gateway
      - type: x-threat-categories
        value: prompt-injection,supply-chain,data-loss,content-safety,agentic-risk
  - aid: guardrails:promptarmor
    name: PromptArmor
    description: >-
      Third-party AI risk and assurance platform spanning TPRM, InfoSec, GRC,
      and Privacy. Assesses risk across 26 vectors aligned with OWASP LLM Top 10,
      NIST AI RMF, and MITRE Atlas. Best known for research on indirect prompt
      injection in Claude for Excel, Google Antigravity, Slack AI, and Writer.com.
    humanURL: https://promptarmor.com/
    tags:
      - GRC
      - Indirect Prompt Injection
      - MITRE Atlas
      - NIST AI RMF
      - OWASP LLM Top 10
      - Third-Party Vendor
      - TPRM
    properties:
      - type: Homepage
        url: https://promptarmor.com/
      - type: x-deployment
        value: Platform
      - type: x-threat-categories
        value: indirect-prompt-injection,data-exfiltration,vendor-risk
  - aid: guardrails:wallarm-ai-security
    name: Wallarm AI Security
    description: >-
      Wallarm extends its API security platform with AI-specific protections
      covering the OWASP LLM Top 10, prompt injection, and abuse of LLM-backed
      API endpoints. Deploys as a sidecar, reverse proxy, or in-line API gateway.
    humanURL: https://www.wallarm.com/product/ai-security
    tags:
      - API Security
      - OWASP LLM Top 10
      - Prompt Injection
      - Reverse Proxy
      - Third-Party Vendor
      - Wallarm
    properties:
      - type: Homepage
        url: https://www.wallarm.com/product/ai-security
      - type: Documentation
        url: https://docs.wallarm.com/
      - type: x-deployment
        value: API Gateway
      - type: x-threat-categories
        value: prompt-injection,api-abuse,owasp-llm-top10
  - aid: guardrails:confident-ai
    name: Confident AI
    description: >-
      AI quality and safety platform behind the open-source DeepEval evaluation
      framework and DeepTeam red-teaming framework. Provides LLM evaluation,
      observability, and red-teaming for OWASP Top 10 for Agentic Applications
      risks including goal hijack, instruction injection, tool misuse, and PII
      leakage.
    humanURL: https://www.confident-ai.com/
    tags:
      - Agentic AI
      - DeepEval
      - DeepTeam
      - Evaluation
      - Observability
      - Open Source
      - Red Teaming
    properties:
      - type: Homepage
        url: https://www.confident-ai.com/
      - type: GitHub
        url: https://github.com/confident-ai/deepeval
      - type: Documentation
        url: https://docs.confident-ai.com/
      - type: x-deployment
        value: SDK
      - type: x-threat-categories
        value: agent-goal-hijack,tool-misuse,prompt-injection,pii,bias
  - aid: guardrails:layerup-ai
    name: Layerup AI
    description: >-
      Note Layerup AI has pivoted toward agentic AI for insurance workflows
      (claims, underwriting). Earlier positioning included LLM guardrails and
      PII redaction; the current product surface focuses on insurance automation
      rather than horizontal AI guardrails. Catalogued for historical context.
    humanURL: https://uselayerup.com/
    tags:
      - Agentic AI
      - Historical
      - Insurance
      - Pivot
    properties:
      - type: Homepage
        url: https://uselayerup.com/
      - type: x-deployment
        value: Platform
      - type: x-threat-categories
        value: historical
common:
  - type: Repository
    url: https://github.com/api-evangelist/guardrails
  - type: JSONSchema
    url: json-schema/guardrail-policy-schema.json
  - type: JSONSchema
    url: json-schema/guardrail-violation-schema.json
  - type: JSONLD
    url: json-ld/guardrails-context.jsonld
  - type: Vocabulary
    url: vocabulary/guardrails-vocabulary.yml
  - type: Examples
    url: examples/
maintainers:
  - FN: Kin Lane
    email: [email protected]
    url: https://apievangelist.com