Vendor
Configurable safety layer for filtering, PII handling, hallucination reduction, and policy controls.
AWS
Proprietary
SaaSSOC 2License medium
Control plane layer
AI Guardrails & Agent Security
Runtime guardrails, content safety, and agent security posture controls.
Compare cloud guardrails with independent safety and agent-security vendors on certifications, data residency, and deployment ownership — full details on each tool's page.
- In dataset
- 12
Also covered by suites listed elsewhere: Amazon Bedrock AgentCore (Registry & management), Databricks Unity Gateway (AI & MCP gateways), Kong AI Gateway (AI & MCP gateways), LiteLLM (AI & MCP gateways), Microsoft Agent 365 (Registry & management), Portkey AI Gateway (AI & MCP gateways), ServiceNow AI Control Tower (Registry & management)
12 matching tools
More filters
Major vendor tools
Cloud-native vendor offerings are shown first here, before the broader open source and third-party landscape below.
Azure AI Content Safety
Microsoft
Content moderation and safety service with configurable severity scoring and on-prem options.
Azure
Proprietary
SaaSOn-premSOC 2License medium
Google Model Armor
Model-agnostic prompt and response safety screening layer across Google enterprise infrastructure.
GCP
Proprietary
SaaSSOC 2License medium
Vendor
Purview posture tool that discovers AI apps and agents, flags oversharing and risky prompts, and applies DLP to AI use.
Azure
Proprietary
SaaSLicense medium
Prisma AIRS
Palo Alto Networks
Palo Alto Networks AI security platform for agent discovery, model scanning, red teaming, posture and runtime protection.
Proprietary
SaaSHybridLicense medium
| Attribute | Azure AI Content Safety | Amazon Bedrock Guardrails | Google Model Armor |
|---|---|---|---|
| Description | Content moderation with configurable severity scoring | Configurable safeguards for filtering, PII, hallucinations | Model-agnostic safety screening for prompts and responses |
| Risks covered | Toxicity, prompt injection, hallucination, copyright | Content, prompt attacks, PII, hallucination, malicious code | Prompt injection, PII, toxicity, malicious URLs |
| Pricing | Pay-per-text-unit; free tier | $0.15/1K text units after the recent Bedrock Guardrails price cut | Pay-as-you-go |
| Unique capability | On-premises containers + on-device deployment path | Automated Reasoning for formal-policy style checks | Apigee + GKE + Security Command Center integration path |
| Integration | REST API, Python/C#/Java SDKs | AWS SDK, Bedrock API, AgentCore | REST API, Vertex AI, Apigee, GKE |
Filtered open source and third-party tools
7 matching non-vendor tools in this filtered view.
Guardrails AI
Guardrails AI
Validation-focused guardrails framework with a large validator ecosystem and freemium model.
AWSAzureGCP
7,404Apache 2.0
Self-hostedLicense low
Lakera Guard
Check Point
Commercial LLM security layer integrating into broader cloud security offerings after acquisition.
Proprietary
SaaSSelf-hostedSOC 2License high
LLM Guard
Palo Alto Networks (Protect AI)
Open source LLM safety-filtering toolkit; the repository was archived in July 2026 and is no longer maintained.
AWSAzureGCP
3,205MIT
Self-hostedLicense low
Archived — Repository archived by the owner on July 9, 2026; the project and its Hugging Face models are no longer under active development.
NeMo Guardrails
NVIDIA
Open source guardrails framework for LLM and agent safety with a dedicated Colang DSL.
AWSAzureGCP
7,109Apache 2.0
Self-hostedLicense low
NVIDIA NemoClaw
NVIDIA
Open-source reference stack that runs always-on agents like OpenClaw inside NVIDIA OpenShell sandboxes with policy and managed inference.
Apache 2.0
Self-hostedLicense low
Rebuff
Palo Alto Networks
Prompt injection defense project archived in May 2025 and not recommended for new projects.
AWSAzureGCP
1,522Apache 2.0
Self-hostedLicense low
Archived — Archived May 16, 2025. Do not recommend for new projects.
Zenity
Zenity
AI agent security platform combining agent posture management, runtime detection and response, and inline blocking.
Proprietary
SOC 2License medium
Important notes
Warning
LLM Guard: Repository archived by the owner on July 9, 2026; the project and its Hugging Face models are no longer under active development.
Warning
Microsoft Purview Data Security Posture Management: The standalone DSPM for AI experience is now labelled "classic" and replaced by the unified DSPM, which receives new features (Microsoft Learn, checked 2026-09-30).
Warning
NeMo Guardrails: Repository moved from NVIDIA/NeMo-Guardrails to NVIDIA-NeMo/Guardrails.
Warning
NVIDIA NemoClaw: Alpha project per the README (checked 2026-09-30): maintainers respond on a best-effort basis; no tagged releases yet.
Warning
Prisma AIRS: Prisma AIRS 3.0 announced 2026-03-23; its AI Agent Gateway was in limited preview at launch.
Warning
Rebuff: Archived May 16, 2025. Do not recommend for new projects.
Recent updates
See auto-detected changes2026-10-09
Google Model Armor
Google Model Armor
Model Armor improved prompt-injection and jailbreak protection for Workspace data such as emails, docs and files, on US and EU multi-region endpoints with filter version v3 or later.
Source2026-09-16
NeMo Guardrails
NeMo Guardrails
NeMo Guardrails 0.24.1 fixed benchmark concurrency measurement, Nemotron 3.5 response parsing, server overload handling, and combined-config loading.
Source2026-08-14
Guardrails AI
Guardrails AI
Guardrails AI 0.11.0 deprecated its Hub CLI and private validator registry as the project moved validator distribution to public PyPI.
Source2026-07-09
LLM Guard
LLM Guard
Protect AI archived the LLM Guard repository; the project and its Hugging Face models are no longer under active development.
Source2026-07-01
NeMo Guardrails
NeMo Guardrails
NeMo Guardrails 0.23.0 expanded IORails with streaming and non-streaming tool-call validation, added a standalone /v1/checks validation endpoint, and improved OpenTelemetry observability for production guardrail pipelines.
SourcePopular comparisons in this category
Read these tools next to one another across every tracked governance dimension.
Frequently asked questions
What does the guardrails & agent security hub cover?
Runtime guardrails, content safety filters, model policy controls, and agent security posture tools — the layer that decides whether a prompt, output, or agent action is safe right now. Fleet-wide agent inventory lives under Agent Registry & Management and agent identities under Agent Identity & Access, both in the same control plane layer. Each per-tool page records data residency, audit logging, SOC 2 / ISO 27001 / ISO 42001, and EU AI Act role.
Are EU AI Act risk-tier assignments tracked per tool?
Yes. Every tool carries an EU AI Act role (prohibited / high-risk / limited-risk / minimal-risk / not-applicable / unknown) on its governance posture, sourced from the vendor's published positioning or set to 'unknown' when no public statement exists.
Which governance tools support on-prem or sovereign deployment?
Self-hostable governance frameworks (NeMo Guardrails, Guardrails AI, Llama Guard, Granite Guardian) run wherever the host runtime is supported. Cloud-native guardrails (Bedrock Guardrails, AI Content Safety, Model Armor) inherit the parent platform's residency surface.
Explore adjacent hubs
Compare the active category against the platform foundation layer, the cross-category updates feed, and the sourcing/contribution guide.
Platforms
Review Microsoft Foundry, Amazon Bedrock, and Gemini Enterprise Agent Platform as the foundation layer behind this category.
Weekly updates
Check the market-intelligence feed for high-impact moves, plus the expandable full log for releases, deprecations, acquisitions, and other notable changes.
About and contribution rules
See sourcing standards, contribution rules, and project scope before adding or updating tracked tools.