{
  "listings": [
    {
      "slug": "amazon-bedrock-guardrails",
      "name": "Amazon Bedrock Guardrails",
      "company": "Amazon Web Services, Inc.",
      "grade": "A",
      "fit": "caveat",
      "why": "Mask mode replaces PII with its type ({NAME}, {EMAIL}), with UK, US and Canadian identifiers and custom regex, but nothing restores values, and the docs say tool-call arguments, tool results and tool definitions are neither blocked nor masked.",
      "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-sensitive-filters.html",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/amazon-bedrock-guardrails"
    },
    {
      "slug": "google-model-armor",
      "name": "Google Cloud Model Armor",
      "company": "Google LLC",
      "grade": "BB",
      "fit": "caveat",
      "why": "Advanced Sensitive Data Protection templates can transform, tokenize or redact infoTypes in prompts and responses, but Model Armor's docs show no restore step, basic mode is mostly US infoTypes, and each request is single-turn text.",
      "source": "https://docs.cloud.google.com/model-armor/overview.md.txt",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/google-model-armor"
    },
    {
      "slug": "guardrails-ai",
      "name": "Guardrails AI",
      "company": "Guardrails AI, Inc.",
      "grade": "D",
      "fit": "caveat",
      "why": "detect_pii with on_fail=fix replaces entities in the validated output, but nothing restores them, the model must now run locally since hosted inference ended on 25 August 2026, and acc-47 found no way to pass thresholds over REST.",
      "source": "https://raw.githubusercontent.com/guardrails-ai/guardrails/main/docs/examples/check_for_pii.ipynb",
      "confidence": "low",
      "url": "https://www.anchorterminal.com/tools/guardrails-ai"
    },
    {
      "slug": "lakera-guard",
      "name": "Lakera Guard (Check Point AI Guardrails)",
      "company": "Check Point Software Technologies Ltd.",
      "grade": "B",
      "fit": "caveat",
      "why": "payload=true returns the type and location of PII and custom regex matches, but the caller does the masking and nothing restores values; address and phone detectors are US formats only.",
      "source": "https://docs.lakera.ai/docs/api/guard.md",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/lakera-guard"
    },
    {
      "slug": "microsoft-presidio",
      "name": "Presidio",
      "company": "Data Privacy Stack (community organisation, no legal entity stated)",
      "grade": "B",
      "fit": "caveat",
      "why": "Typed replace, mask, hash and encrypt operators with a DeanonymizeEngine to restore, and pseudonymisation samples; but acc-12 saw an unsupported entity dropped without error, acc-11 entities split across stream deltas, and many country recognisers are disabled by default.",
      "source": "https://presidio.dataprivacystack.org/anonymizer/",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/microsoft-presidio"
    },
    {
      "slug": "nemo-guardrails",
      "name": "NVIDIA NeMo Guardrails",
      "company": "NVIDIA Corporation",
      "grade": "BB",
      "fit": "caveat",
      "why": "GLiNER, Presidio and Private AI rails mask PII with label placeholders ([FIRST_NAME]) on input, output and retrieval, but nothing restores values, and acc-41 notes built-in rails inspect message text only, so tool-call arguments are not masked.",
      "source": "https://docs.nvidia.com/nemo/guardrails/latest/configure-guardrails/guardrail-catalog/pii-detection.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": "Contains PII masks input with typed placeholders such as \u003cEMAIL_ADDRESS\u003e via Presidio, but masking in the output stage is not supported, nothing restores values, and acc-38 saw emails, card numbers and Korean phone numbers missed.",
      "source": "https://openai.github.io/openai-guardrails-python/ref/checks/pii/",
      "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": "DLP detections return prompt_masked_data and response_masked_data with offsets, but the spec has no restore operation, and acc-46 saw a plugin fail every scan and log a warning instead of blocking.",
      "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": "agentgate",
      "name": "AgentGate",
      "company": "Mohit Singh (individual)",
      "grade": "D",
      "fit": "no",
      "why": "Scores tool calls; argument values are redacted only in its own audit log (args_redacted), and nothing masks or restores personal data in prompts.",
      "source": "https://github.com/mohitdebian/AgentGate/blob/main/docs/API_REFERENCE.md",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/agentgate"
    },
    {
      "slug": "azure-ai-content-safety",
      "name": "Azure AI Content Safety (Prompt Shields)",
      "company": "Microsoft Corporation",
      "grade": "B",
      "fit": "no",
      "why": "Content Safety's functions are Prompt Shields, harm categories, protected material, groundedness and custom categories; no PII detection or masking is among them.",
      "source": "https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview",
      "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": "no",
      "why": "PII, PCI and PHI rules with customisable entity lists return is_safe and rule classifications; the pages read show no entity locations, masking or restore.",
      "source": "https://raw.githubusercontent.com/cisco-ai-defense/ai-defense-python-sdk/main/README.md",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/cisco-ai-defense-inspection"
    },
    {
      "slug": "granite-guardian",
      "name": "Granite Guardian",
      "company": "International Business Machines Corporation",
      "grade": "B",
      "fit": "no",
      "why": "A yes or no judge over criteria; no entity locations, masking or restore.",
      "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": "S7 Privacy can flag a response as unsafe, but the model returns only safe or unsafe and category codes; it does not locate, mask or restore personal data.",
      "source": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama-guard-4",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/llama-guard"
    },
    {
      "slug": "llamafirewall",
      "name": "LlamaFirewall",
      "company": "Meta Platforms, Inc.",
      "grade": "D",
      "fit": "no",
      "why": "Scanners are Prompt Guard 2, AlignmentCheck, CodeShield and regex, plus an experimental PII scanner that detects but does not mask or restore.",
      "source": "https://meta-llama.github.io/PurpleLlama/LlamaFirewall/docs/documentation/about-llamafirewall",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/llamafirewall"
    },
    {
      "slug": "mistral-moderation",
      "name": "Mistral Moderation API",
      "company": "Mistral AI (RCS Paris 952 418 325)",
      "grade": "B",
      "fit": "no",
      "why": "A pii category score says whether text holds personal data, but nothing locates, masks or restores it.",
      "source": "https://docs.mistral.ai/studio/safety-moderation.md",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/mistral-moderation"
    },
    {
      "slug": "openai-moderation",
      "name": "OpenAI Moderation API",
      "company": "OpenAI OpCo, LLC",
      "grade": "BB",
      "fit": "no",
      "why": "Harm-category classifier only; no PII detection, masking or restore.",
      "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": "Classifies four harm types in text and image policies in ShieldGemma 2; no personal-data detection or masking.",
      "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": 13,
    "after": [
      "inference.tool-loop",
      "agent-observability.gateway-logging"
    ],
    "area": "models",
    "asks": [
      "strip personal data before this goes to the model",
      "redact names and emails from these logs",
      "anonymise customer records before the LLM sees them",
      "mask PII in prompts and restore it in the answer"
    ],
    "before": [
      "mailbox-access.find-and-read",
      "web-scrapers.page-to-markdown",
      "document-extraction.read-file"
    ],
    "blockers": [
      "entity type for the user's country not shipped or disabled by default",
      "masking breaks when the response is streamed, so placeholders reach the user",
      "detector error returns an empty result and the request goes out unmasked"
    ],
    "capabilities": [
      "guard.pii",
      "guard.self-host"
    ],
    "category": "guardrails",
    "categoryName": "Guardrails \u0026 safety filters",
    "checked": "2026-11-15",
    "companies": {
      "fit": 0,
      "fitWithLimit": 8,
      "tagged": 18
    },
    "competition": "none",
    "fits": null,
    "fitsWithLimit": [
      "amazon-bedrock-guardrails",
      "google-model-armor",
      "guardrails-ai",
      "lakera-guard",
      "microsoft-presidio",
      "nemo-guardrails",
      "openai-guardrails",
      "prisma-airs"
    ],
    "id": "guardrails.redact-pii",
    "listings": {
      "caveat": 8,
      "no": 9,
      "notRead": 0,
      "tagged": 20,
      "unchecked": 0,
      "yes": 0
    },
    "name": "Mask personal data before it reaches a hosted model, and put it back after",
    "status": "draft",
    "success": "Every configured entity type in the input is replaced by a typed placeholder before the call, the response comes back with the original values restored, tool-call arguments in the history are masked too, and an unsupported entity or a detector failure is reported rather than skipped silently.",
    "url": "https://www.anchorterminal.com/api/v1/use-cases/guardrails.redact-pii.json",
    "weight": 22,
    "weightBasis": "13 of 50 accounts, the largest group: Presidio issues (country identifiers, streaming, silent drops), LiteLLM Presidio-guardrail issues, two Show HN builders and one OpenAI Guardrails issue; share of 58 account-to-use-case links, rounded"
  }
}
