AI is now a "basic expectation" for the public service. Governance has to be the other one.

A few decades ago, when I worked in a Government Ministry, there were a few phrases you heard often.

“Give the Minister options.”

“Keep the Minister off the front page of the Dom.”

And perhaps most importantly:

“Do what the Minister asks for.”

That last one mattered. Government is ultimately a political democratic system. Ministers weigh up a whole range of factors that officials may not see, and then they make decisions. A non-partisan public sector has to understand that.

So when Nicola Willis MP made AI a “basic expectation” for every public agency, I was less interested in whether AI commentators agreed with her, and more interested in what public sector Boards and chief executives should now actually do.

I work in AI governance for sectors where accuracy matters. AI governance is new, and really about how organisations de-risk the adoption of generative AI while still getting the benefits.

If I was invited into the room with public sector Boards and exec teams, this are the governance actions I would start with.

At Board level:

1. Set an AI risk appetite statement

A short, Board-approved statement that says what the organisation will use AI for, what lines it will not cross, and which decisions must always keep a human clearly in charge. This gives staff a boundary. “Use AI” must never become “use AI anywhere!”

2. Put AI permanently on the meeting agenda

Not an occasional paper when a project needs approval. A standing item, with honest reporting on incidents, overrides, near misses, complaints and lessons learned. AI is not IT.

3. Fund AI governance

There needs to be protected budget for testing, assurance, audit and review. Governance is not bureaucracy. It is what lets you move faster without being reckless.

4. Build AI literacy at the top

Directors and executives need enough understanding to interrogate the reporting, challenge the assumptions and ask the hard questions. A Board that doesn't understand the strengths and limitations of AI is not governing AI.

If the executive team were also in the room, I would add:

5. Be serious about ROI

The best AI opportunity is not always the easiest, shiniest or most visible one. Executives must be the ones to decide where limited investments should go. That means looking for genuine return, not just activity and adoption.

6. Create a prohibited-use list

Be explicit about where AI must not make, or materially shape, a decision on its own. Draw the line now, before someone under deadline pressure draws it for you.

7. Name the accountable human

Every AI-assisted service or decision process needs a named human owner. Guardrails, mandatory checks and workflow controls all help, but accountability itself cannot be delegated. There must be clear points where a person reads, understands and approves before something reaches a member of the public. I love using Hertz as a case study in my workshops, human override is essential!!

8. Capture knowledge before people leave

If AI is being used partly because headcount is reducing, then knowledge capture is not optional. It is both respectful as well as practical as you cannot ask AI to support a process without the context needed.

If the Minister was in the room, I would suggest two across-government protections:

9. A shared AI incident register across agencies

One agency’s bad day should become everyone’s lesson!

10. A clear right to human review

When AI has materially shaped a decision about a person, they should be able to ask for a human to look again. That is basic fairness and how to protect public trust. One bad story can drain confidence from the whole programme.

The Minister has set the pace and we need to follow those three phrases I learnt at the Ministry of Defence. We need to support our public sector leaders to turn a mandated rollout into something better than what we have today: faster and fairer services, supported by AI, and delivered by people who remain firmly in charge.

This was also published on LinkedIn Pulse: https://www.linkedin.com/pulse/ai-now-basic-expectation-public-service-governance-has-matt-ensor-5xs8e