In short
- Train in three levels: AI literacy for everyone, skills by role, then a small group of champions.
- Build safeguards into every level. Do not leave them to a module at the end.
- Move from one-off courses to a pathway. Singapore’s Institute of Digital Government, launched in September 2026, offers structured learning pathways and mandatory foundational modules for public officers.
- Measure what people do at work, not who attended.
- The gap is real. In ACCA’s 2025 survey of more than 10,000 accountancy and finance professionals, 38% said they were not confident in their AI knowledge and 32% said their organisation provided chances to learn AI skills.
Most civil service AI training starts with a tool demonstration and ends with a certificate. People enjoy it, and little changes. A better approach is a sequence: who learns what, in what order, with the rules built in. This article sets out a three-level sequence you can adapt to an agency or a whole service. It is written for heads of HR and learning, digital and data leaders, and the sponsors who fund the training.
Why the order matters
Training without a sequence tends to fail in the same ways. Leaders approve uses they do not understand. Officers use public tools in ways that put data at risk. Specialists wait for rules that never arrive.
The skills gap is wide. ACCA’s Global Talent Trends 2025 survey drew more than 10,000 accountancy and finance professionals in 175 countries. Half of respondents (50%) were concerned they were not developing the skills the future workplace requires. In all, 38% said they were not confident in their current knowledge of AI, and 44% cited insufficient opportunities to learn about AI as a key barrier. Separately, fewer than a third (32%) said their organisation was providing opportunities to learn AI-related skills. The respondents are finance professionals, not civil servants, so treat the figures as a signal, not a measure of any one civil service.
The policy direction is similar. In January 2024, an ERIA researcher writing on ASEAN public sector jobs named reskilling and upskilling the public workforce as the key policy response. And on 18 September 2026, Singapore launched its Institute of Digital Government, set up by the Ministry of Digital Development and Information with the Civil Service College. It offers structured learning pathways across proficiency levels, requires all public officers to complete annual foundational modules on cybersecurity, data protection and AI literacy, and runs agency-level “AI Sprints” for hands-on practice. Its stated aim is to equip more than 150,000 public officers with digital, data and AI skills by 2028.
Level 1: AI literacy for everyone
Everyone who works in the agency needs the same basics, because everyone will meet AI, whether through a tool the agency buys or one they find themselves.
- What to cover: what generative AI is and is not, where it is strong and where it fails, how to check its output, what must never be pasted into a public tool, when a person must review the result, and the agency’s rules.
- Format: short modules, ideally built around realistic scenarios rather than slides.
- Outcome: every officer can explain the basics and the rules, and knows who to ask.
Level 2: skills by role
Once the basics are in place, each role needs something different. Give people practice on their own tasks.
| Role | What they need | Example output |
|---|---|---|
| Senior leaders | How to decide: what to approve, what questions to ask, who is accountable. | An AI use approval checklist. |
| Policy officers | Using AI to research and draft with verification, and judging effects on policy. | A policy note with a verification log. |
| Analysts and data teams | Data quality, evaluating AI outputs, designing workflows. | An evaluated use case. |
| Front-line and service staff | Using approved tools in daily work, handling errors, escalating. | A revised service script. |
| HR and learning teams | Mapping skills, redesigning jobs, designing training. | A reskilling plan. |
| Procurement, legal and risk | Assessing vendors, contracts and privacy. | A vendor assessment checklist. |
Level 3: champions
Choose a small group, for example one or two per team, to coach their peers, test tools in safe settings and report problems back to whoever governs AI use. Give them protected time. Champions turn a course into a habit, because colleagues learn from people they work beside.
Build safeguards into every level
Safeguards are a skill to practise, not a policy to read. At each level, include:
- Privacy: what data may and may not be used.
- Accuracy: how to check claims, sources and numbers.
- Fairness: how bias can enter and how to spot it.
- Human review: which decisions always need a person.
- Records: how to note where AI was used.
Formats that work
- Short modules that fit into a working week.
- Scenario simulations where people handle realistic cases.
- Practice on real tasks, with feedback.
- Peer review, for example reviewing a colleague’s work sample.
- A capstone project tied to a real workplace challenge.
- Assessment-based certification, so skills are checked, not assumed.
How to measure it
Attendance is not impact. Use four levels:
| Level | Question | Example measure |
|---|---|---|
| Reaction | Did people find it useful? | A short survey right after. |
| Learning | What did they learn? | A before and after check. |
| Behaviour | What do they do differently at work? | A manager check or work sample at three months. |
| Results | What changed for the service? | A measure agreed before the training, reviewed at twelve months. |
A 12-month sequence
This is a suggested plan to adapt.
| Months | What happens |
|---|---|
| 1 to 2 | Baseline and rules: run a skills survey, issue interim guidance on AI use, brief senior leaders. |
| 3 to 4 | Level 1: roll out AI literacy to everyone. |
| 5 to 8 | Level 2: start with leaders and analysts, then policy officers, front line and support teams. |
| 9 to 12 | Level 3: name and support champions. Measure behaviour change. Adjust the programme. |
Common pitfalls
- Training before rules. People learn tools they are not yet allowed to use.
- One course for everyone. Leaders and front-line staff need different things.
- No time to practise. Skills fade without use.
- No follow-up. A single session rarely changes habits.
- Counting attendance. It tells you who turned up, not what changed.
An example from our work
For a public education body, we built a programme around three parts: AI literacy, classroom application and safeguarding. The same logic carries over to other public-sector roles. Only the examples change. Read the anonymised case study.
Frequently asked questions
What AI skills should civil servants learn first?
Start with AI literacy: what AI can and cannot do, how to check its output, what data must stay out of public tools, and the agency's rules. Then add skills for each role, so that leaders, analysts, policy officers and front-line staff each practise on their own tasks.
How long does AI training for civil servants take?
Literacy can be short, in modules of under an hour. Role-based skills take longer and work best spread over several months with practice on real tasks. Champions build up over a year. The 12-month sequence above shows one way to plan it, which you can adapt to your agency.
Do all officers need technical training?
No. Most officers need literacy, rules and role-specific practice. Only data and technology specialists need deep technical skills. Leaders need to know what to approve and what to ask, and front-line staff need to know how to use approved tools, handle errors and escalate problems.
What to do next
Start with the baseline: a short skills survey and an interim rule on AI use. Then plan Level 1. If you would like help designing or delivering the sequence, discuss a project with Blak Rice. See our training programmes and government work.
Sources
- ACCA, Global Talent Trends 2025, executive summary: accaglobal.com. The 32% figure is from ACCA’s headline findings, as reported by PrimeGlobal.
- Ministry of Digital Development and Information, “New Institute of Digital Government to Build an AI-Ready Public Service”: mddi.gov.sg
- GovInsider, “Singapore launches Institute of Digital Government to scale digital and AI skills among public officers”: govinsider.asia
- ERIA, “AI on ASEAN’s public sector jobs: a helping hand or a threat to employment?” (30 January 2024): eria.org