# AI Governance Toolkit

> Govern AI without slowing it down: an AI register with risk tiers, 32 lifecycle controls, a quarterly board report, policy, standards and an audit programme.

Source: https://ciohelm.com/toolkits/ai-governance-toolkit · Area: Govern · Price and packs: https://ciohelm.com/llms.txt

Most companies already use dozens of AI systems, and most of them were bought, not built: features switched on in software they already pay for. Few can say what they use, who owns it, how risky it is or whether the controls are in place. This kit gives you a governance system that fits the risk: one register, four tiers, four checkpoints and 32 controls, with the policy, standards and board reporting around them, all filled in for a worked example.

Excel, PowerPoint and Word files. Yours to use and adapt inside your organisation.

## 15 working files, not a slide pack to read

Every file is editable and comes with a worked example for one fictional company, so you start from something, not from a blank page. Workbooks calculate as you type.

### AI governance workbook (Excel)

20 sheets that run AI governance: setup, the AI register, risk classification with prohibited and high-risk triggers and a factor score, the 32-control library, assurance system by system with coverage and a watch list, impact assessments, conditions and exceptions, provider assurance by use, incidents, risks, obligations, training, KPIs, maturity, roadmap, decision rights and the log. A dashboard pulls it together.

### AI governance framework (PowerPoint)

14 slides to get the framework adopted: why now, the framework on a page, eight principles, four tiers, classification, four checkpoints, 32 controls, generative AI and agents, who decides, bought AI, the starting point and a twelve-month roadmap.

### Quarterly AI governance report (PowerPoint)

11 slides for the board or executive committee, worked through: the register, control coverage, high tier systems, incidents, risks, measures, people and providers, maturity and decisions.

### AI governance charter and decision rights (Word)

The forums, who sits on them, what each decides, and a decision rights matrix for twelve decisions.

### AI policy (Word)

Principles, risk appetite, scope, prohibited uses, the rules for every employee and how exceptions are handled.

### AI risk classification standard (Word)

Prohibited and high-risk triggers, six scored factors, four tiers and their review cycles, with examples.

### AI impact assessment (Word)

The assessment for high tier systems: purpose, people affected, data, fairness, oversight, risks and sign-off.

### AI lifecycle and control standard (Word)

The four checkpoints and the 32 controls, which tier needs each one and the evidence it takes, mapped to NIST AI RMF functions and ISO/IEC 42001 Annex A topics.

### AI system record (Word)

The record every system keeps: owner, purpose, data, provider, tier, controls, monitoring and changes.

### Generative AI and AI agents standard (Word)

Extra rules for generated content, prompts and data, and for agents that act: limits, permissions and human approval.

### Third-party AI assurance (Word)

The questionnaire and contract clauses for AI you buy, and how to keep watching a provider after signature.

### AI testing, monitoring and human oversight standard (Word)

What to test before release, what to monitor in operation, incident severity levels, and how people stay in charge.

### AI obligations map (Word)

How to map the laws, contracts and commitments that apply to your AI, with an owner for each.

### AI governance audit programme (Word)

Audit objectives and test steps for each component, for internal audit or a second line review.

### Governing AI without slowing it down (Guide)

How to start, scale the controls to the risk, run the first quarter and avoid the common mistakes.

## AI governance workbook, sheet by sheet

The workbook carries the method: scores, choices and plans calculate as you type, and the dashboard and deck figures come from it. Click any image to see it full size.

**Dashboard**: Systems by tier, control coverage, the watch list, reviews due, incidents, training and providers.

**Register**: Every AI system with its owner, provider, stage, type and calculated tier and next review.

**Classification**: Prohibited and high-risk triggers and six scored factors that set each tier.

**Assurance**: The controls due for each system, their status and coverage against the target.

## Control that follows the risk

Every AI system, bought or built, is in one register. Clear triggers and a factor score set its tier, and the tier decides which of the 32 controls apply and how often it is reviewed. Coverage is calculated system by system, so the board sees which systems fall short and why. In the sample, 24 systems become three high tier systems watched closely, and most low-risk tools pass with a light touch.

## The method, step by step

### Month 1 · Find

Name the owner and the forums, adopt the policy, and list every AI system in use, including features in software you already buy.

### Months 2-3 · Classify

Classify each system, stop anything prohibited, and assess the high tier systems first.

### Months 3-6 · Control

Put the checkpoints into intake, design and release, close the control gaps and assure your providers.

### Every quarter · Report

Report coverage, incidents, risks and measures to the board, review systems on their cycle and raise the maturity.

### Who it is for

- CIOs, CDOs and CTOs asked to show the board that AI is under control
- Risk, compliance, legal and privacy leads building an AI governance function
- Heads of data and AI who want clear rules that speed delivery up
- Advisers setting up AI governance for a client (one licence per client organisation)

### What you need

Excel, PowerPoint and Word, or compatible software. A first version takes about three months; the quarterly cycle a few days.

### Free guide on this topic

- [AI governance: 12 questions boards should ask](https://ciohelm.com/guides/ai-governance-board-questions)

## Questions about the kit

**Does it make us compliant with the EU AI Act or ISO/IEC 42001?**

No toolkit can do that on its own. The framework is built to be consistent with them and maps each control to NIST AI RMF functions and ISO management system themes, so it gives you a sound base. Confirm which obligations apply to you with your legal advisers.

**We mostly buy AI rather than build it. Is it still relevant?**

Yes, more so. The register, classification and controls cover AI inside the software you buy, and the third-party assurance document gives you the questions and contract clauses for providers.

**Is it too heavy for a smaller company?**

No. Low tier systems need 12 controls, mostly light ones. Start with the register, the policy and the classification, and add the rest as your use of AI grows.

**Are the systems and providers in the example real?**

No. The manufacturer, its 24 systems and Provider A to Q are fictional. Replace them with your own.

**Is it legal or financial advice?**

No. The templates are practical starting points for management decisions. Have policies and contracts reviewed against the law that applies to your company.

**Who can use it?**

You and your colleagues, inside your own organisation. Advisers need one licence per client organisation. Please do not resell or republish it.

## Working files from an experienced CIO, ready for your next meeting.
