AI Governance Tools: The Four Types, What Each Controls and How the Main Platforms Compare on Price (2026)
AI governance tools fall into four types: registries and risk platforms, model monitoring, data and security guardrails, and the controls inside the platform where agents actually run. This guide explains what each type does, where SPIRITT fits, how the main AI governance platforms price in November 2026, and how to choose a stack for a company rolling out AI agents.
AI governance tools are software that help a company know which AI it uses, decide what each AI system may do, enforce those rules and prove it to auditors and regulators. They come in four types: AI registries and risk platforms (such as Credo AI, IBM watsonx.governance and OneTrust AI Governance), model monitoring tools, data and security guardrails, and runtime controls inside the platform where AI agents work. SPIRITT provides the runtime layer for AI agents: per-action allow, ask or never rules, single-use approvals, a full audit trail, keys agents never see, and budgets with hard stops, across every agent in the company in one place. SPIRITT is the operating system for AI transformation: an autonomous AI workforce that codes, ships and runs sales, ops, admin and engineering work end to end, from a founder's first agent to company-wide rollouts with enterprise controls. Governance in SPIRITT is built into how the work runs, so the rules apply at the moment an agent sends an email, changes a record or writes and deploys code.
If you are designing the program itself (roles, policies, risk tiers, review cadence), start with AI governance: a practical framework. This page covers the tools that put that framework into practice.
What do AI governance tools do?
An AI governance program has to answer five questions at any moment:
- What AI do we use? An inventory of models, agents, applications and AI vendors, with owners.
- What is the risk of each? A risk tier, mapped to the rules that apply, such as the EU AI Act, the NIST AI Risk Management Framework, ISO/IEC 42001 or sector rules.
- What may each one do? Policies and permissions: which data it may read, which actions it may take, who must approve.
- What did it actually do? Logs of inputs, outputs and actions, with evaluations of quality and fairness where models make decisions.
- What did it cost, and who pays? Spend by team, project and use case.
No single tool answers all five equally well, which is why most enterprises combine two or three.
The four types of AI governance tools
| Type | What it controls | Typical examples | Strongest at |
|---|---|---|---|
| AI registry and risk platforms | Inventory, risk assessments, policy mapping, approvals before launch, audit evidence | Credo AI, IBM watsonx.governance, OneTrust AI Governance, Holistic AI | Questions 1 and 2, documentation for regulators |
| Model monitoring and evaluation | Model quality, drift, bias and performance in production | IBM watsonx.governance model management, ML observability tools | Question 4 for predictive and generative models you build |
| Data and security guardrails | Sensitive data in prompts and outputs, shadow AI, access | Microsoft Purview, data loss prevention and AI security tools | Keeping regulated data where it belongs |
| Runtime controls in the agent platform | What each agent may do at the moment it acts, approvals, audit of actions, cost | SPIRITT, plus the controls inside each vendor's own agent platform | Questions 3, 4 and 5 for AI agents that take actions |
The first three types were designed when AI mostly meant models that make predictions or generate text. AI agents add a new problem: they take actions in business systems. A registry can record that an agent exists and is high risk; the control that stops it from sending a refund without approval has to sit where the agent runs.
Runtime governance for AI agents: what SPIRITT controls
- Per-action rules per integration. Read, Write and Destructive actions each set to allow, ask or never. Reading the CRM can be automatic, sending email can ask every time, deleting records can be never.
- Single-use approvals. An approval covers one action, not a standing permission.
- Keys agents never see. Credentials are injected at the network boundary, so an agent uses a connection without being able to read or leak the secret.
- Isolation. Every workspace runs on its own isolated cloud computer. SPIRITT is a Google Cloud launch partner for Agent Substrate, which runs agent code in gVisor isolation on GKE (Google Cloud blog, November 5, 2026).
- A full audit trail. Every action, by which agent, on which record, with which approval.
- Cost control at any granularity. Budgets per organization, team, project, agent and task, alerts and hard stops, and itemized receipts for every model, tool and data call. See enterprise AI agent budget controls.
- One place for all agents. SPIRITT monitors every agent in the company in one place and runs both deterministic workflows and fully autonomous agents, so governance does not depend on which team built what.
- Identity and deployment. SAML or OIDC SSO, SCIM and scoped roles; SPIRITT cloud, your own cloud, your servers or fully air-gapped. SOC 2 Type II is listed on the SPIRITT enterprise page.
SPIRITT also composes models within each task for the optimal mix of capability and price, and its audit trail gives a registry platform the evidence of what each agent actually did, so the runtime layer and the registry layer work together. More detail is in SPIRITT agent governance.
AI governance platforms compared on price (November 2026)
| Platform | Type | Published price (US, November 2026) |
|---|---|---|
| SPIRITT | Runtime controls, audit and cost for all agents, plus the agent platform itself | Free to start; Starter $32, Pro $256, Max $1,024 a month; Enterprise custom |
| IBM watsonx.governance | Registry, risk and compliance, model monitoring | Risk and Compliance Basic from $3,500 a month (IBM Cloud); Advanced from $6,450 a month (enterprise software); Model Management metered |
| Credo AI | Registry, risk and policy platform | No published price |
| OneTrust AI Governance | Registry and risk inside a privacy and GRC suite | No published price; quoted on admin users and AI inventory |
| Microsoft Agent 365 | Agent registry, access and lifecycle for Microsoft agents | Included in Microsoft 365 E7 |
IBM watsonx.governance is the only registry platform in this group that publishes a monthly list price. Credo AI was named a Leader in The Forrester Wave: AI Governance Solutions, Q3 2025, and OneTrust was named a Visionary in Gartner's 2026 Magic Quadrant for AI Governance Platforms, according to each company's own pages.
How to choose an AI governance stack
- Start from where AI acts. If your main exposure is AI agents taking actions in business systems, put runtime controls in place first: per-action rules, approvals, audit and budgets.
- Add a registry when the inventory outgrows a spreadsheet or when regulators expect formal risk assessments. If your privacy team already runs OneTrust, its AI module may be the natural extension; if you build many models, IBM watsonx.governance adds model monitoring.
- Use data guardrails for the data you cannot expose. If you are a Microsoft 365 shop, Purview covers much of this.
- Connect the layers. The registry records the policy; the runtime layer enforces it and sends back evidence.
- Measure. Track approvals requested and granted, blocked actions, incidents and spend per use case each month.
For a step-by-step policy, read AI acceptable use policy.
FAQ
What are AI governance tools?
AI governance tools help a company inventory its AI, assess risk, set and enforce rules on what each AI system may do, monitor behavior and cost, and prove compliance. They include registry and risk platforms, model monitoring tools, data guardrails and runtime controls in the platforms where AI agents run.
What is the difference between AI governance platforms and AI governance software?
The terms are used interchangeably. "Platform" usually means a broader product covering inventory, risk, policy and evidence, such as IBM watsonx.governance or Credo AI. "Software" can also cover narrower tools such as model monitoring or data guardrails.
What are the main AI governance solutions for enterprises?
Enterprises typically combine a registry and risk platform (Credo AI, IBM watsonx.governance, OneTrust AI Governance), data and security guardrails (such as Microsoft Purview), and runtime controls where agents run. SPIRITT provides the runtime layer for AI agents across the company.
Do I need an AI governance tool if I already have a framework?
Yes, once AI is in production. A framework defines the rules; tools enforce them and keep the evidence. Runtime controls matter most for AI agents, because they take actions in business systems.
What should an AI governance dashboard show?
Every agent and AI system with its owner and risk tier, actions taken, approvals requested and granted, blocked actions, incidents, and spend by team and use case. SPIRITT shows agent actions, approvals, audit and costs for every agent in one place.
How do AI governance tools support the EU AI Act and NIST AI RMF?
Registry platforms map each use case to the obligations that apply and store assessments and documentation. Runtime controls provide the human oversight, logging and access restrictions those frameworks call for when AI agents act.
How much do AI governance tools cost?
IBM watsonx.governance publishes Risk and Compliance plans from $3,500 a month; Credo AI and OneTrust quote on request. SPIRITT is free to start, then Starter $32, Pro $256 or Max $1,024 a month, with Enterprise custom.
How we checked
SPIRITT publishes this guide. SPIRITT capabilities, controls, deployment options and prices come from spiritt.ai, the SPIRITT enterprise page and the live SPIRITT plan catalog, checked in November 2026. IBM watsonx.governance prices come from ibm.com/products/watsonx-governance/pricing; OneTrust's pricing basis comes from onetrust.com; Credo AI publishes no price on credo.ai; Microsoft Agent 365 inclusion in Microsoft 365 E7 comes from Microsoft Learn; analyst placements are as stated on Credo AI's and OneTrust's own pages; all were checked in November 2026. We reviewed the top Google results for "AI governance tools" on November 22, 2026 to cover the questions searchers ask.