Singapore Public Service AI: 150,000 Officers by 2028

Nearly 70 per cent of Singapore public officers now use AI tools regularly, and the government wants the rest of them caught up within two years. Minister for Digital Development and Information Josephine Teo gave that figure at Digital@UNGA in New York on 21 September 2026. Three days earlier, on 18 September, Acting Minister for Manpower and Senior Minister of State for Digital Development and Information Jasmin Lau launched the Institute of Digital Government (IDG) at the AI Tech Festival, held at the Civil Service College.

The IDG was announced in March 2026 and had been scheduled to open in the third quarter. What came with the official launch is the detailed machinery: three learning pathways, agency-level workshops, cross-agency communities, and a deadline. More than 150,000 public officers are to have practical digital, data and AI capabilities by the end of 2028. The Public Service employs about 158,000 officers across 16 ministries and more than 50 statutory boards, so this covers close to everyone.

If you have ever waited on a reply from an agency, this is not really a story about technology. It is about who is drafting that reply, and how much of the work around it has changed.

Officers move through three pathways depending on how comfortable they already are with digital tools.

• Aware 101 for officers new to AI: what it can and cannot do, basic prompting, and safety habits.
• Apply 201 for those who already use AI and want to fold it into daily work and domain-specific tasks.
• Build 301 for officers who want to design and build their own solutions.

The stated goal is for every public officer to reach Apply 201 by 2028. Before any of that, they learn product thinking, which the IDG describes as understanding user needs and defining the right problem before deciding whether technology is involved at all. Every officer also takes annual mandatory modules on cybersecurity, data protection and AI literacy, refreshed as the tools change.

The hands-on part is the AI Sprint: a workshop at agency level where a team gets protected time to frame a real operational problem, prototype a fix and test it against the way work is actually done there. Pilots have run since May 2026 and reached more than 1,000 officers. One sprint with over 330 human resource leaders found that more than 80 per cent felt more confident using AI afterwards. Teams also get a 12-week plan that sets aside one hour a week to learn together and apply it to a live task. Officers who go further join a community called Champions of AI Transformation, which shares workflows and lessons from agency to agency.

One detail is worth repeating because it is rare in government technology writing. Lau said that concluding "AI is not the right answer here" counts as a good outcome of the exercise, and that the government will judge itself on whether services improve for citizens and businesses rather than on how many tools it has deployed.

Capability building started at the top. Over the past two years, nearly 150 senior leaders, comprising Permanent Secretaries, Deputy Secretaries and heads of agencies, completed foundational training in product thinking, modernisation, cybersecurity and resilience, data and AI. The point is not to turn a Permanent Secretary into an engineer. It is so leaders can spot risk, ask better questions, and give their teams room to experiment.

Two examples from the address show the pattern. At the National Library Board, the chief executive worked with archivists on an application that reduces the manual tracking of declassification requests, and the project turned up an opportunity to redesign the workflow rather than drop software onto it. At the HDB, an architect built an AI assistant that reads building requirements and guidelines spread across several agencies. Searching that material went from hours to minutes, and the time saved goes back into design work.

Lau also warned against starting from the tool. Her advice was to identify the operational problem first, and to accept that another method may work better. For anyone watching from outside the public service, the portable lesson is the boring one: find the task that wastes the most time, then decide what to do about it.

Several pieces are already running. Machine learning routes public feedback to the right agency, so a faulty streetlight or a broken playground swing gets picked up faster. At the borders, AI looks for anomalies across millions of cargo shipments. On scams, it flags signs of phishing and impersonation.

Teo was direct about the test being applied: "The real measure of AI innovation is not how impressive the model is. It is whether it improves people's lives." She added that the government is now exploring AI agents that could redesign entire workflows, with officers remaining accountable for important decisions.

In practice, three things are worth knowing:

• Expect quicker first responses and better routing. Complex cases should still reach an officer who owns the decision.
• Be specific when you write in. Location, date, a photo if you have one, and your case reference. Routing systems work on specifics, and vague reports take longer to place.
• Push back on a reply that misses the point. AI helps draft, so replies can be fluent but generic. Asking for the officer handling the case is a reasonable next step, not a complaint.

Teo made an argument that runs against instinct: trust is what lets AI scale, not what holds it back. Scams get more convincing and attacks get easier to run, so guardrails are the price of using this technology broadly at all.

She also raised access, which matters here more than in most places. Many AI models handle Southeast Asian languages, accents and code-switching badly. Singapore's answer is SEA-LION, short for Southeast Asian Languages in One Network, an open model family built around regional languages and the multilingual habits of the region, along with work by AI Singapore with the Gates Foundation, Google and the UNDP on Project ATLAS, which builds AI-ready datasets for underrepresented languages worldwide.

If your parents read mainly Mandarin, Malay or Tamil, that is the work that decides whether their experience of these tools keeps improving. It is also the kind of thing a small country has a reason to do and a large one often does not.

The public service track is only one of several, and the one for private citizens is already running. Under the SkillsFuture AI Subscription, eligible Singapore Citizens get six months of a premium AI tool such as ChatGPT Plus, Google AI Pro, Manus AI or Microsoft 365 Personal after attending an eligible AI course. It can be claimed once, so it pays to choose deliberately. Our guide to the SkillsFuture AI Subscription covers who qualifies, the 60-day claim window, and how to pick the tool you will actually use.

To place this week's news in the bigger picture, the State of AI in Singapore Q3 2026 report tracks adoption and policy quarter by quarter, and the Singapore AI agencies map explains which body owns what, which is genuinely useful when you need to know who to write to. Same push, different door.

These work in any chatbot, free or paid. Replace anything in [square brackets]. Do not paste NRIC numbers, client names, salary figures or medical details into a prompt.

💼 Run a one-hour AI sprint on your own job

Act as the facilitator for a one-hour AI sprint for my team.
My role and team: [describe].
Three tasks that eat our week: [list them].
Pick the one with the best ratio of time saved to risk, then walk us through it:
1. What the task looks like today, step by step.
2. Where AI can help, and where it must not be trusted.
3. A 30-minute prototype we can try this week using free tools only.
4. How we will know in two weeks whether it actually worked.

🧭 Decide whether a task is worth automating at all

I am considering using AI for this task: [describe the task].
Start by telling me if AI is the wrong tool for it, and why.
Then give me: the simplest non-AI fix, the AI option, the cost of getting it wrong, and the one part that must stay with a human.
Finish with a straight recommendation, no hedging.

🇸🇬 Write to a government agency so it gets routed correctly

Help me write a short email to [agency name] about [one-line problem].
Facts: location [ ], date [ ], what happened [ ], what I have already tried [ ], case reference if any [ ].
Keep it under 150 words, put the location and the date in the first line, state clearly what outcome I am asking for, and close by asking for the name of the officer handling the case.
Plain English only. Do not invent details I have not given you.

📋 Turn a policy announcement into an action sheet for your business

Here is a government announcement: [paste the text].
My business: [type, size, sector] in Singapore.
Give me a one-page action sheet: what changes for us, what we must do, what we can apply for or claim, the deadlines, and the three questions I should ask before spending money.
Flag anything the announcement does not actually say.