---
title: What are the categories of AI governance tools?
url: https://opencontroller.lyzr.com/blog/categories-of-ai-governance-tools
published: 2026-11-15
updated: 2026-11-15
author: Lyzr
---

# What are the categories of AI governance tools?

**AI governance tools fall into six main categories: AI security posture management (AI-SPM), LLM gateways, GRC and AI policy modules, model risk and evaluation tools, AI data-loss prevention and browser security, and agent control planes. They differ in what they can see and what they can enforce. Most enterprises end up with two or three, so the question is which combination covers your agents without gaps or overlap.**

The market labels overlap, and many products stretch across categories. Judge each tool by two questions: what does it observe, and where can it act?

## What does each category do?

| Category | What it sees | What it can enforce | Typical owner |
|---|---|---|---|
| AI-SPM | Cloud AI resources, configs, permissions, data exposure | Findings and remediation guidance; little at run time | Cloud security |
| LLM gateway | Model requests and responses that route through it | Routing, rate limits, caching, redaction, model allowlists | Platform engineering |
| GRC / AI policy module | Questionnaires, risk assessments, policies, attestations | Workflow and approvals on paper; no run-time control | Risk and compliance |
| Model risk and evaluation | Model and agent behaviour under test | Release gates based on eval scores | Data science / model risk |
| AI DLP and browser security | Employee prompts and uploads to AI apps in the browser or network | Block or redact sensitive data in user sessions | Security operations |
| Agent control plane | Agents, versions, tools, identities, traffic and cost across platforms | Registry rules, policy, block, quarantine, kill across the estate | Security + AI platform |

## Where does each category fall short?

Each one is good at its slice and blind outside it:

- **AI-SPM** sees cloud configuration but not agents built on SaaS platforms or running on laptops. It reports, but rarely stops a live agent. See [AI-SPM vs an AI agent control plane](/blog/ai-spm-vs-ai-agent-control-plane).
- **LLM gateways** see only traffic routed through them. Agents that call models directly, or act through tools without model traffic, are invisible. Compare them in [LLM gateway vs agent gateway vs control plane](/blog/llm-gateway-vs-agent-gateway-vs-control-plane).
- **GRC modules** hold the policy and the risk register, but they rely on people filling in forms. They cannot check whether an agent behaves as declared. See [AI governance platform vs GRC tool](/blog/ai-governance-platform-vs-grc-tool).
- **Model risk tools** test before release. They often stop at the release gate, with no view of production traffic.
- **AI DLP and browser security** protect employees using chat apps. They do not govern autonomous agents running server-side.
- **Agent control planes** cover the agent estate, but they depend on connectors and traffic paths. Registration does not equal run-time control; how deep control goes depends on how each agent is connected.

## How do you choose a combination?

Start from your agents, not from the categories.

1. **List where agents run.** Cloud AI services, SaaS agent platforms, enterprise chat workspaces, developer laptops, custom code.
2. **List the controls you need per location.** Inventory, identity, tool limits, run-time block, cost attribution, evidence.
3. **Map categories to the gaps.** For each location and control, mark which category covers it.
4. **Remove overlap.** Two tools doing inventory create two inventories that disagree.
5. **Decide the system of record.** One place should hold the authoritative list of agents and owners. Other tools feed it.

A common pattern: a GRC module holds policy and the risk register; an agent control plane holds the live inventory and enforces at run time; an LLM gateway (often part of the control plane) handles model routing and cost; AI DLP covers employee chat use.

## What questions expose a tool's real category?

Ask vendors these in a demo:

- "Show me an agent you discovered that nobody registered."
- "Stop this agent now, and show me where that action took effect."
- "Show the cost of this agent version last week."
- "Show the evidence record for this control, mapped to a framework clause."

A tool that can answer only one or two is a point product. That is fine, as long as you know which gaps remain. Use the [evaluation checklist](/blog/ai-agent-governance-platform-evaluation-checklist) for a fuller list.

## How Opencontroller does this

Opencontroller is an agent control plane. It auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, connects to third-party agent platforms and enterprise chat workspaces, and keeps a governed registry. It includes an LLM Gateway with cost attribution per agent and version, an Agent Gateway, Posture, evaluations and rogue-agent handling, with policy mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. Opencontroller Edge extends control to developer devices. See the [control plane comparison](/compare/controlplane) and the [capability reference](/capability-reference), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Do I need a separate tool for each category?**
No. Several categories are often combined in one product, especially gateways inside control planes. Map your controls first, then buy to fill gaps.

**Which category should be the system of record for AI agents?**
The one that sees the most agents with live data, usually an agent control plane. GRC tools can consume that inventory for risk and compliance workflows.

**Is an LLM gateway enough for AI agent governance?**
Not on its own. It sees only traffic routed through it and does not cover agent identity, tools, discovery or estate-wide quarantine.

## Key facts

Six categories of AI governance tools: AI security posture management (AI-SPM: cloud configs and permissions, findings), LLM gateways (routing, rate limits, caching, redaction for traffic routed through them), GRC/AI policy modules (risk registers, assessments, approvals), model risk and evaluation tools (pre-release testing, release gates), AI DLP and browser security (employee prompts and uploads), and agent control planes (agents, tools, identities, traffic, cost; block, quarantine, kill across the estate). Choose by mapping where agents run and needed controls, remove overlap, pick one system of record. Opencontroller (by Lyzr) is an agent control plane with discovery across AWS, Azure, Google Cloud, a governed registry, LLM and Agent Gateways, Posture, evals and rogue-agent handling. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [What is the difference between AI-SPM and an AI agent control plane?](https://opencontroller.lyzr.com/blog/ai-spm-vs-ai-agent-control-plane.md)
- [Do you need an AI agent governance platform if you already have a GRC tool?](https://opencontroller.lyzr.com/blog/ai-governance-platform-vs-grc-tool.md)
- [Should you buy AI agent governance as separate modules or as one suite?](https://opencontroller.lyzr.com/blog/buy-ai-governance-as-modules-or-suite.md)
- [More on Buying and comparison](https://opencontroller.lyzr.com/blog/cluster/buying-and-comparison)
