{
  "listings": [
    {
      "slug": "amazon-bedrock-guardrails",
      "name": "Amazon Bedrock Guardrails",
      "company": "Amazon Web Services, Inc.",
      "grade": "A",
      "fit": "caveat",
      "why": "ApplyGuardrail screens text and image content blocks with per-policy assessments and errors as HTTP codes, but tool-use fields are skipped (acc-18) and a gateway integration dropped images from the payload (acc-19).",
      "source": "https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ApplyGuardrail.html",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/amazon-bedrock-guardrails"
    },
    {
      "slug": "azure-ai-content-safety",
      "name": "Azure AI Content Safety (Prompt Shields)",
      "company": "Microsoft Corporation",
      "grade": "B",
      "fit": "caveat",
      "why": "shieldPrompt takes one userPrompt and up to five documents, not a message history, within 10K characters; no fail-mode or not-run status is documented, and document attack detection is off by default in Foundry.",
      "source": "https://learn.microsoft.com/en-us/azure/ai-services/content-safety/region-availability",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/azure-ai-content-safety"
    },
    {
      "slug": "cisco-ai-defense-inspection",
      "name": "Cisco AI Defense Inspection API",
      "company": "Cisco Systems, Inc.",
      "grade": "B",
      "fit": "caveat",
      "why": "Chat inspection takes the whole messages array, HTTP inspection covers raw requests, and the SDK has a gateway mode with enforce or monitor; but the API never blocks itself and no fail-mode or not-run status was found in what was read.",
      "source": "https://developer.cisco.com/docs/ai-defense-inspection/getting-started/",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/cisco-ai-defense-inspection"
    },
    {
      "slug": "google-model-armor",
      "name": "Google Cloud Model Armor",
      "company": "Google LLC",
      "grade": "BB",
      "fit": "caveat",
      "why": "Each prompt or response is inspected as a single-turn request with no conversation history; a skipped or failed check is reported (EXECUTION_SKIPPED, invocation_result FAILURE), but acc-20 and acc-21 show Responses traffic unscreened and batches failing in a gateway.",
      "source": "https://docs.cloud.google.com/model-armor/overview.md.txt",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/google-model-armor"
    },
    {
      "slug": "lakera-guard",
      "name": "Lakera Guard (Check Point AI Guardrails)",
      "company": "Check Point Software Technologies Ltd.",
      "grade": "B",
      "fit": "caveat",
      "why": "Only the last interaction is screened, earlier turns are context; content type text only, so images are not checked. acc-22 saw a 403 with no detector named and acc-24 a v1 body sent to v2 that blocked all traffic.",
      "source": "https://docs.lakera.ai/docs/api/guard.md",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/lakera-guard"
    },
    {
      "slug": "mistral-moderation",
      "name": "Mistral Moderation API",
      "company": "Mistral AI (RCS Paris 952 418 325)",
      "grade": "B",
      "fit": "caveat",
      "why": "Inline guardrails with per-category thresholds, action block and block_on_error cover Mistral's chat, conversations and agents only; the conversational endpoint classifies the last turn, and text only.",
      "source": "https://docs.mistral.ai/studio/safety-moderation.md",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/mistral-moderation"
    },
    {
      "slug": "nemo-guardrails",
      "name": "NVIDIA NeMo Guardrails",
      "company": "NVIDIA Corporation",
      "grade": "BB",
      "fit": "caveat",
      "why": "The server fronts any model with an OpenAI-compatible endpoint and /v1/checks, and tool rails fail closed, but acc-03 found only the last user message checked and acc-04 that a cut-off verdict passes as safe.",
      "source": "https://docs.nvidia.com/nemo/guardrails/latest/configure-guardrails/guardrail-catalog/tool-calling.md",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/nemo-guardrails"
    },
    {
      "slug": "openai-guardrails",
      "name": "OpenAI Guardrails",
      "company": "OpenAI (as named in the LICENSE copyright line and the PyPI author field)",
      "grade": "BB",
      "fit": "caveat",
      "why": "A client wrapper rather than a gateway; checks fail open unless raise_guardrail_errors=True, and the prompt injection check's own table gives a 1,481 ms median for the default model, far above the 150 ms bar.",
      "source": "https://openai.github.io/openai-guardrails-python/quickstart/",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/openai-guardrails"
    },
    {
      "slug": "prisma-airs",
      "name": "Prisma AIRS AI Runtime Security API",
      "company": "Palo Alto Networks, Inc.",
      "grade": "C",
      "fit": "caveat",
      "why": "Every response carries action allow or block, a category, and errors for detection services that failed or timed out; but only the last element of contents is scanned, earlier turns are context, and content is text only.",
      "source": "https://raw.githubusercontent.com/PaloAltoNetworks/pan.dev/master/openapi-specs/prisma-airs/scan/scan-service_latest.yaml",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/prisma-airs"
    },
    {
      "slug": "guardrails-ai",
      "name": "Guardrails AI",
      "company": "Guardrails AI, Inc.",
      "grade": "D",
      "fit": "unchecked",
      "why": "The docs site returned 404, and the repository pages read do not say which request routes the API server screens, whether history is checked, or how a validator that fails to run is reported.",
      "confidence": "low",
      "url": "https://www.anchorterminal.com/tools/guardrails-ai"
    },
    {
      "slug": "agentgate",
      "name": "AgentGate",
      "company": "Mohit Singh (individual)",
      "grade": "D",
      "fit": "no",
      "why": "Sits between an agent and its tools, assessing tool invocations; it does not screen chat, Responses, Messages or batch model traffic.",
      "source": "https://github.com/mohitdebian/AgentGate/blob/main/docs/API_REFERENCE.md",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/agentgate"
    },
    {
      "slug": "granite-guardian",
      "name": "Granite Guardian",
      "company": "International Business Machines Corporation",
      "grade": "B",
      "fit": "no",
      "why": "An 8B judge model the owner runs, one criterion per call; not a gateway or policy layer.",
      "source": "https://huggingface.co/ibm-granite/granite-guardian-4.1-8b",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/granite-guardian"
    },
    {
      "slug": "llama-guard",
      "name": "Llama Guard 4",
      "company": "Meta Platforms, Inc.",
      "grade": "D",
      "fit": "no",
      "why": "A classifier model the owner hosts, not a policy layer; no injection or PII checks, and the card sends readers to Prompt Guard 2 for prompt attacks.",
      "source": "https://raw.githubusercontent.com/meta-llama/PurpleLlama/main/Llama-Guard4/12B/MODEL_CARD.md",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/llama-guard"
    },
    {
      "slug": "llamafirewall",
      "name": "LlamaFirewall",
      "company": "Meta Platforms, Inc.",
      "grade": "D",
      "fit": "no",
      "why": "An in-process Python library that scans messages the agent hands it; it is not a proxy and has no route coverage, moderation or PII masking.",
      "source": "https://meta-llama.github.io/PurpleLlama/LlamaFirewall/docs/documentation/about-llamafirewall",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/llamafirewall"
    },
    {
      "slug": "microsoft-presidio",
      "name": "Presidio",
      "company": "Data Privacy Stack (community organisation, no legal entity stated)",
      "grade": "B",
      "fit": "no",
      "why": "A PII detection and anonymisation library with no gateway, injection or moderation checks of its own; route coverage belongs to the proxy that calls it, and acc-15, acc-16 and acc-43 show LiteLLM's Presidio path missing tool arguments and streams.",
      "source": "https://presidio.dataprivacystack.org/anonymizer/",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/microsoft-presidio"
    },
    {
      "slug": "openai-moderation",
      "name": "OpenAI Moderation API",
      "company": "OpenAI OpCo, LLC",
      "grade": "BB",
      "fit": "no",
      "why": "A harm classifier, not a policy layer: no injection, PII or conversation-wide checks, and inline moderation on a streamed generation arrives only after the full output.",
      "source": "https://developers.openai.com/api/docs/guides/moderation",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/openai-moderation"
    },
    {
      "slug": "shieldgemma",
      "name": "ShieldGemma",
      "company": "Google LLC (RDAP registrant of gemma.ai; the Gemma Terms of Use are Google's and the cards name Google as author)",
      "grade": "C",
      "fit": "no",
      "why": "An open-weight classifier the owner runs, one policy per call; no injection or PII checks and no gateway.",
      "source": "https://gemma.ai/models/shieldgemma/model-card-v1.md",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/shieldgemma"
    }
  ],
  "marks": "yes: the docs show it does the job as the success test says. caveat: it does the job with a limit an agent should know. no: it doesn't do this job. unchecked: we couldn't read it.",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-16",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.8",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "useCase": {
    "accounts": 11,
    "after": [
      "agent-observability.gateway-logging",
      "human-in-the-loop.audit-decisions"
    ],
    "area": "models",
    "asks": [
      "put a guardrail in front of all our model traffic",
      "make every team's calls go through the same checks",
      "add Bedrock or Model Armor screening to our LiteLLM proxy"
    ],
    "before": [
      "inference.failover-routing"
    ],
    "blockers": [
      "guardrail reads only chat-completions messages, so Responses or batch traffic passes unscreened",
      "API version change makes every check fail and the gateway blocks all traffic",
      "image parts dropped from the guardrail payload while still sent to the model"
    ],
    "capabilities": [
      "guard.policy",
      "guard.injection",
      "guard.pii",
      "guard.moderation"
    ],
    "category": "guardrails",
    "categoryName": "Guardrails \u0026 safety filters",
    "checked": "2026-11-15",
    "companies": {
      "fit": 0,
      "fitWithLimit": 9,
      "tagged": 25
    },
    "competition": "none",
    "fits": null,
    "fitsWithLimit": [
      "amazon-bedrock-guardrails",
      "azure-ai-content-safety",
      "cisco-ai-defense-inspection",
      "google-model-armor",
      "lakera-guard",
      "mistral-moderation",
      "nemo-guardrails",
      "openai-guardrails",
      "prisma-airs"
    ],
    "id": "guardrails.gateway-enforcement",
    "listings": {
      "caveat": 9,
      "no": 7,
      "notRead": 0,
      "tagged": 27,
      "unchecked": 1,
      "yes": 0
    },
    "name": "Enforce one guardrail policy on every request through a gateway",
    "status": "draft",
    "success": "Every request route the gateway serves (chat, Responses, Messages, batch, streaming) is screened, the whole conversation is checked rather than only the last message, a check that did not run is logged as not run, and a block returns a reason the caller can act on. Measured: the check adds under 150 ms at the median for a 2,000-token request.",
    "url": "https://www.anchorterminal.com/api/v1/use-cases/guardrails.gateway-enforcement.json",
    "weight": 19,
    "weightBasis": "11 of 50 accounts: LiteLLM, Otari, Future AGI and OpenClaw-plugin issues where guardrails skipped a route, dropped images, failed open or blocked everything, plus a NeMo gateway integrator's latency benchmark"
  }
}
