AI governance for mid-sized companies, without the year-long programme
Your staff are already using ChatGPT, Claude and Copilot. Your vendors have switched AI on inside software you already pay for. Some of it is starting to act on its own. AI governance is how you get from that to a position you can defend to a board, a customer or a regulator, in weeks rather than a year.
Haden McAlpine · CISSP · ISO/IEC 27001 Lead Auditor · 25 years in IT, cyber and infrastructure
What AI governance actually has to produce
Strip away the frameworks and every serious approach asks for the same four things. If you have these, you can answer almost any question about your AI use with evidence rather than reassurance.
An AI register
Every AI system and AI feature in use, including the ones built into Microsoft 365, the CRM, HR and helpdesk tools, each with a named owner and a risk rating.
An AI use policy
Which tools are approved, what data can go where, who owns the output, and what needs a human sign-off. Short enough that people read it.
Controls that run
Screening for new AI features, risk assessment per use case, data loss prevention and logging for AI tools, and monitoring after go-live.
A board pack
A short paper: where AI is used, the risks, the controls, and the decisions the board needs to make. Updated as the register changes.
Which framework fits
The frameworks overlap far more than their marketing suggests. Pick the one your buyers and regulators recognise, and the same register, policy and controls will serve the others.
| Framework | What it is | Use it when |
|---|---|---|
| AI6 (Australia) | The National AI Centre’s Guidance for AI Adoption, October 2025. Voluntary. Six essential practices, with free templates. | You operate in Australia, or sell to Australian customers who ask how you govern AI. |
| ISO/IEC 42001:2023 | The international, certifiable standard for an AI management system. Same structure as ISO 27001. | A customer, tender or regulator wants a certificate, or you already run ISO 27001. |
| NIST AI RMF 1.0 | The US National Institute of Standards and Technology’s voluntary AI risk framework, with a Generative AI profile. | You sell into the US, or want a detailed risk vocabulary to work from. |
| EU AI Act | Binding EU regulation, in force since 1 August 2024 and applying in stages, with the main high-risk obligations from 2 December 2027. | You place AI systems on the EU market or use them in the EU, including UK companies with EU customers. |
AI6’s Implementation practices were written to align with ISO/IEC 42001 and the NIST AI RMF, and the National AI Centre publishes a crosswalk between them. Work done for one counts towards the others.
Agentic AI changes the risk
A chat window produces text. An agent takes actions: it sends emails, updates records and calls other systems through connectors like MCP servers. That’s where most of the new risk is, and where most governance programmes are thinnest.
- Least privilege for agents. An agent gets the narrowest access that does the job, scoped per tool, not the permissions of whoever set it up.
- Human approval before high-stakes actions. Anything that moves money, changes customer records or sends external messages waits for a person.
- Every tool call logged. If you can’t reconstruct what the agent did and why, you can’t investigate it.
- A tested off switch. Someone named can stop the agent, and has actually tried it.
- Prompt injection treated as a given. Agents that read email, web pages or documents will meet instructions planted in them. Design for it, using the OWASP Top 10 for LLM Applications as the checklist.
I run my own practice on governed AI agents with exactly these controls, so this is what I’ve had to live with, not just what I recommend.
Services by region
Australia & New Zealand →
AI governance aligned to AI6 and ISO/IEC 42001, plus the Privacy Act automated-decision rules from 10 December 2026 and NZ’s IPP 3A.
UK & Europe →
EU AI Act readiness and AI governance built on your existing ISO 27001 system: inventory, risk classification, policy and controls.
Reading
Common questions
What is AI governance?
AI governance is the set of decisions, owners and controls that make sure an organisation's use of AI is deliberate, safe and accountable. In practice, for a mid-sized company, that means an AI register, an AI use policy, risk screening and controls for each use case, and a short report the board can approve.
Which AI governance framework should we use?
Use the one your buyers and regulators recognise. In Australia that is the Guidance for AI Adoption (AI6). In the UK and EU, the EU AI Act sets binding obligations for anyone placing AI on the EU market. Internationally, ISO/IEC 42001 is the certifiable standard and the NIST AI RMF is a widely used risk framework. They overlap heavily, so the same register, policy and controls serve all of them.
Do we need ISO/IEC 42001 certification?
Usually not yet. Certification makes sense when a customer, tender or regulator asks for it. The governance work comes first either way, and if you already hold ISO 27001 the step to 42001 is much smaller.
How is governing agentic AI different?
An AI agent can take actions, not just produce text: send emails, change records or call other systems through connectors such as MCP servers. That means least-privilege access for the agent, human approval before high-stakes actions, logging of every tool call, and a tested way to stop it. Australia's AI6 guidance explicitly asks organisations to treat agentic AI as its own risk category.