What skills and support do public servants need to deploy AI safely and confidently?
When I sat down with Viscount Camrose, Shadow Minister for DSIT, during AI Public Sector Week, we discussed what it truly takes for public servants to deploy AI safely and confidently. Personally, I came away from our discussion with a conviction I had not quite articulated before: the most important ingredient in government AI adoption is not the algorithm. It is the human holding it.
That might sound obvious. But spending time in this conversation made me realise how often we still talk about AI in government as though the hard part is procurement, or infrastructure, or compliance. It is not. The hard part is people: their confidence, their judgement, and the organisational culture surrounding them.
Here is what stayed with me.
The prize is cross-government, but we keep playing department by department
It is relatively easy to find a use case within a single team. Automating casework, drafting correspondence, summarising lengthy reports. These are real gains, and they matter. But they are not where the transformative value lives.
The bigger opportunity is systemic. An AI-powered signal in one department that routes attention across several others. Coordinated delivery that does not require every ministry to reinvent the wheel. That kind of impact demands inter-departmental coordination, shared standards, and someone with genuine authority to drive it. Without that leadership at cabinet level, with real mandate and budget, we are left with pockets of innovation rather than public service transformation.
Scrutiny is not the enemy of progress
The government faces something most organisations do not: judicial reviews, parliamentary oversight, FOI requests, and constant media attention. It is tempting to treat this scrutiny as a brake on innovation. I do not think that is the right frame.
Scrutiny underpins public trust. The challenge is designing AI governance that enables learning alongside accountability, not instead of it. The risk of overcorrecting toward caution is that teams stall before they ever reach the stage where they can improve and scale. Transparency matters, but it cannot be the only instrument in the toolkit.
The accountability rule is simple, and we should say it louder
The human is always responsible for what the AI output is used for.
This is not a technicality. It is the principle that unlocks confidence. When people understand that AI is a tool and not a decision-maker, and that accountability sits firmly with them, they can engage with it more honestly. They can check outputs, flag problems, and own the results. The moment we blur that line, we do not just create legal risk. We erode the public trust that the government depends on.
AI literacy is not the same as AI engineering
When we talk about skilling up the civil service, the instinct is sometimes to reach for technical training. Data science, model evaluation, prompt architecture. Some of that is necessary in specialist roles. But it is not the core need.
What most civil servants actually need is a different combination: the confidence to use AI tools in everyday work, the judgement to evaluate what comes back, the awareness to spot bias or inappropriate content, and the instinct to know when something looks wrong. That is AI literacy. It is practical, it is teachable, and far more achievable than turning every policy official into a machine learning engineer.
The rarest skill is spotting opportunity across the whole system
What struck me most in my conversation with Viscount Camrose is that the scarcest capability in government is not technical at all. It is the ability to look across departments, understand how different services connect, and identify where AI could improve outcomes end to end.
The civil servants who will drive genuine transformation are not necessarily those with the deepest technical knowledge. They are those who combine curiosity about what AI can do with a broad understanding of how government actually works. That combination, more than any individual skill, is what turns AI into lasting public value.
Success should be visible to citizens, not just measurable internally
Five years from now, the question should not be “did we deploy AI?” It should be: do people feel that services are faster, fairer, and more reliable? Has the time AI freed up been reinvested in doing more, not simply in doing less? A “do more” mindset matters enormously here. Productivity gains that shrink budgets without improving outcomes are not transformation. They are just efficiency dressed up as ambition.
What I took away from this conversation is that AI in government is as much a question of organisational design and human judgement as it is of technology. The tools are ready. The harder work, building the confidence, the culture, and the coordination to use them well, that is still in front of us.
Watch the full conversation here: https://aipsweek.digileaders.com/talks/what-skills-and-support-do-public-servants-need-to-deploy-ai-safely-and-confidently/
Read More Workforce & AI Skills
