Building AI skills in your teams

Why skills decide the outcome

The same AI tool can save time in one office and cause harm in another. The difference is almost always the people using it: whether they know what it is good for, how to check its output and when not to use it at all. Buying tools without building skills is one of the most common and expensive mistakes in public sector AI. As a senior officer you control the budget, time and permission that skills development needs.

Shadow AI: it is probably already happening

Staff in most organisations are already using free AI tools on personal phones or browsers, often with good intentions: to draft faster, translate a letter or summarise a long report. This is shadow AI, unofficial use that leadership does not know about and has not approved.

The risk is real. Staff may paste classified material or citizens' personal data into public tools, rely on unchecked output or create records nobody can trace. But banning AI outright rarely works. It tends to push use further out of sight.

A better approach:

• Acknowledge it. Assume some use is happening and say so openly, without blame.
• Find out. Ask teams, anonymously if needed, which tools they use and for what.
• Offer a safe route. Provide approved tools or clear rules for when public tools may be used with public, non-sensitive material.
• Draw firm lines. Make it clear that classified material and personal data must never go into unapproved tools.

Training that fits the role

Different staff need different things. A short, role-based approach works better than one long course for everyone:

• All staff: what AI is, what it gets wrong, what must never be shared and how to check output. The operational officers pathway covers this ground.
• Regular users: writing good instructions, checking output systematically and recording AI use properly.
• Managers: setting rules for their teams, reviewing AI-assisted work and spotting automation bias.
• Specialists: procurement, legal, data and IT staff need deeper knowledge of contracts, data protection and evaluation.
• Senior leaders: this pathway.

Aim for AI literacy across the organisation: a shared, practical understanding, not deep technical expertise for a few.

Guidance notes

Staff need short, practical written guidance more than long policies. A good guidance note fits on one or two pages and answers:

• which tools are approved and for which tasks
• what material may and may not be used with them
• what checks are required before AI-assisted work is used
• how AI use should be recorded on files
• who to ask when unsure and how to report a problem

Review the note regularly. Tools and risks change quickly. Make sure guidance is consistent with your department's instructions and any government guidance. The module on responsible use in your office shows the kind of rules a section might set.

Safe spaces to practise

People learn AI by using it. Give staff a safe way to do that:

• an approved tool or clear permission to use public tools with public material only
• practice exercises using published documents, not live files
• protected time, even an hour a month, to experiment and share what they learn
• a way to share good prompts and lessons across teams

Some staff will be early enthusiasts. Use them as peer guides, but make sure they also model careful checking, not just speed.

Leading by example

Staff watch what senior officers do. If you ask for AI-assisted briefs, ask how they were checked. If you hear of a mistake, treat it as a chance to improve guidance, not to punish. If you use AI yourself, say so and say how you checked the result. A culture where people feel safe to report problems is the best protection you have.

Try it

Draft a one-page guidance note for an imaginary section on using AI assistants. Cover approved uses, what must never be shared, required checks, recording AI use and who to ask. Use only general principles and leave placeholders where your department's own instructions would apply.

Check yourself

1. What is shadow AI and why does an outright ban rarely work?
2. Name three groups of staff who need different training.
3. What five questions should a guidance note answer?

Answers: (1) Unofficial AI use leadership does not know about; banning tends to push it further out of sight. (2) Any three of: all staff, regular users, managers, specialists, senior leaders. (3) Approved tools and tasks, permitted material, required checks, how to record use and who to ask or report to.