AI governance tools, by the job they do
This field reference tracks AI governance point tools across six lifecycle stages: test, evaluate, monitor, attack, guard, and secure. It currently lists 16 evidence-gated tools across 5 active chapters, backed by 685 dated single-source facts that cleared deterministic and two-judge review.
The toolchain by lifecycle stage
- Test Ch. 01 · Evaluate
- Evaluate Ch. 01 · Evaluate
- Monitor Ch. 02 · Monitor
- Attack Ch. 03 · Red-team
- Guard Ch. 04 · Guard
- Secure Ch. 05 · Posture
Each stage links to the chapter covering point tools built for that job: test and evaluate (Ch. 01), monitor (Ch. 02), attack and red-team (Ch. 03), guard at runtime (Ch. 04), and secure posture (Ch. 05).
Tool index
| # | Tool | Stage | Licence | What it does | Evidence | Last fact |
|---|---|---|---|---|---|---|
| 01 | Amazon Bedrock Guardrails | Guard | Licence unknown | AWS says Amazon Bedrock Guardrails provides six safeguard policies that can be configured for generative AI applications based on use cases and responsible AI policies. | ||
| 02 | Arize AI (AX & Phoenix) | Monitor | Licence unknown | Atlan says Arize is an AI observability and evaluation company. | ||
| 03 | Arthur AI | Evaluate | Licence unknown | aisecuritydirectory.com describes Arthur AI as a US company that monitors and controls production AI systems. | ||
| 04 | Azure AI Content Safety | Guard | Licence unknown | In 2023, TechCrunch reported that Azure AI Content Safety was a productized version of the safety system powering Microsoft's chatbot in Bing and GitHub's Copilot code-generating service. | ||
| 05 | DeepEval (Confident AI) | Evaluate | Licence unknown | Confident AI describes itself as an AI quality platform built by the creators of DeepEval. | ||
| 06 | Fiddler AI | Monitor | Licence unknown | Fiddler lists 100+ pre-defined metrics and support for custom metrics. | ||
| 07 | garak (NVIDIA) | Red-team | Licence unknown | AppSecSanta describes garak as NVIDIA’s open-source LLM vulnerability scanner with 50+ probe modules for prompt injection, jailbreaks, and hallucinations. | ||
| 08 | Giskard | Red-team | Licence unknown | Giskard describes itself as an AI agent evaluation and red teaming platform. | ||
| 09 | HiddenLayer | Posture | Licence unknown | TechCrunch says HiddenLayer makes tools to protect AI models and agents as well as workflows from adversarial attacks and vulnerabilities as well as malicious code injections. | ||
| 10 | Lakera | Guard | Licence unknown | In 2025, Atomico reported that Lakera’s co-founders were David Haber, Matthias Kraft and Mateo Rojas-Carulla. | ||
| 11 | Langfuse | Monitor | Licence unknown | Langfuse describes itself as an open-source AI engineering platform for building, monitoring, and improving LLM applications. | ||
| 12 | LangSmith | Evaluate | Licence unknown | DropAgentic identifies LangSmith Engine as a LangChain tool. | ||
| 13 | Mindgard | Red-team | Licence unknown | Mindgard's GitHub profile describes it as a provider of AI red teaming and security testing solutions. | ||
| 14 | NVIDIA NeMo Guardrails | Guard | Licence unknown | AppSecSanta describes NeMo Guardrails as an open-source toolkit by NVIDIA for adding programmable guardrails to LLM-based conversational applications. | ||
| 15 | Promptfoo | Red-team | Licence unknown | TechCrunch describes Promptfoo as an AI security startup founded in 2024 to protect LLMs from online adversaries. | ||
| 16 | Weights & Biases Weave | Monitor | Licence unknown | GitHub describes Weave as a toolkit for tracing and evaluating AI agents, built by Weights & Biases. |
Point tools versus governance platforms
This reference covers standalone technical point tools that perform a specific governance function—evaluation, observability, red teaming, guardrails, or AI security posture. Broad enterprise compliance suites and agent-runtime control planes are out of scope here.
The ordering rule
Organic order follows search demand plus evidence completeness; sponsored entries are labelled.