Public Service AI / AI news for Malaysia
From the archive · Event date 12 March 2025
Malaysia trained 200 AI change ambassadors. The 445,000-officer rollout depended on what happened next
The Train-the-Trainer programme gave public officers practical Gemini skills and a mandate to teach their departments. The next test was whether 200 trainers could produce safe, measurable adoption across a much larger public service.
In brief
- About 200 Change Ambassadors from ministries and agencies attended a five-day Train-the-Trainer programme in Putrajaya from 3 to 7 March 2025.[1][2]
- They were expected to spread practical Gemini skills inside their departments as part of AI At Work 2.0, which aimed to provide tools and learning support to as many as 445,000 public officers.[1][3]
- The official records describe training and self-reported pilot gains, but they do not establish how many officers later became trained active users or how citizen-facing services changed.[1][3]
A central training cohort was created to spread practical AI skills across government
Malaysia's public-service AI push moved from a large access target to a smaller group of human trainers in March 2025. Around 200 Change Ambassadors from ministries and agencies spent five days at Menara Usahawan in Putrajaya learning how to use generative AI for day-to-day government work.[1][2]
The participants were also described as MyGovUC IT administrators in the National Digital Department. Their job was not simply to finish a workshop. The ministry said they would carry the knowledge back into their respective departments, turning one central training programme into a wider internal learning network.[1]
That distinction is important. Two hundred trained ambassadors are evidence of a trainer cohort; they are not evidence that 445,000 officers were trained, using the tools regularly or delivering better services. Those wider outcomes would require a separate record of multiplication, safe use and measured results.[1][3]
The training focused on ordinary office work, not experimental AI projects
The programme covered four practical functions: generating content, summarising information, analysing data and improving meetings with AI-generated notes. These are familiar knowledge-work tasks, which made the training relevant to policy, research, administration, integrity, human resources and technical teams rather than only software specialists.[1][2]
The ministry highlighted several officers to show how the tools were being considered in real roles. Sheila Mahalingam described automating repetitive work to create more room for strategy. Mohd Fazli Sahari discussed faster access to articles, journals and other sources, while Chor Yee Reen pointed to transcription, translation and summarising minutes and reports.[1]
Those examples make the programme concrete, but they also show why training must include verification. A faster summary can still omit context. A quicker research scan can still cite a weak source. AI-generated meeting notes can still misattribute a decision. The time saved only becomes useful when the officer remains accountable for checking the final output.[1]
The real operating challenge was multiplying 200 trainers across a 445,000-person opportunity
AI At Work 2.0 had been launched on 5 February 2025 with an ambition to make Google Workspace's Gemini tools and learning support available to up to 445,000 public officers. The Change Ambassadors were one mechanism for turning that broad opportunity into usable knowledge inside individual agencies.[1][3]
A Train-the-Trainer model can scale efficiently because each ambassador understands the systems, colleagues and work patterns in a department. It can also become uneven. One agency may provide time, examples and leadership support; another may leave a single administrator to answer questions without an approved workflow or clear escalation route.[1]
The useful public scoreboard would therefore separate four stages: ambassadors trained, officers subsequently trained, monthly active users on approved tasks, and services with verified outcome improvements. Combining those stages into one large access number would make it impossible to tell where adoption was working or stalling.[1][3]
The pilot showed promising self-reported savings, not a nationwide productivity result
The ministry said an earlier pilot involved 270 officers from the National Digital Department. It reported that 91% saw better work quality and saved an average of 3.25 hours per week. That was a useful signal for deciding whether to expand training.[1][3]
The figures were participant-reported pilot results. They were not presented as an independent time study, and they did not show whether checking, correction and rework were fully included. They also cannot be multiplied across 445,000 officers because job types, usage frequency and service risks vary widely between agencies.[1][3]
Malaysia had also launched Public Sector AI Adaptation Guidelines on 27 February. The ministry said those guidelines covered ethical principles, roles, risk management, adoption methods and self-assessment. Training, guardrails and outcome reporting therefore needed to operate together: people need skills, agencies need boundaries, and the public needs evidence that the change improved services safely.[1]
Why Malaysia should care
The programme matters because AI-assisted drafting, analysis and meeting work could touch services used by millions of Malaysians. A credible rollout needs to show not only who received tools, but who learned to use them, where human review remained, and whether services became faster without introducing avoidable errors.
Public officers
The training made common drafting, research, analysis and meeting tasks easier to explore with AI.[1]
Practical move: Keep source links, check every material fact and preserve human approval before an AI-assisted output becomes an official record.
Agency leaders
A trained ambassador can accelerate adoption, but one workshop does not create a safe operating model across an agency.[1]
Practical move: Give ambassadors approved use cases, protected learning time, risk escalation and a dashboard separating training, use and outcomes.
Malaysians using public services
The programme could shorten routine work, but the public record reviewed did not establish nationwide service improvements.[1][3]
Practical move: Look for published service-level measures, error rates and correction routes rather than relying only on access or training totals.
What Malaysians can do now
- Track the multiplication rate from each Change Ambassador to trained officers instead of treating the first 200 as the final adoption result.
- Report active use only for approved workflows and publish time, quality, correction and incident measures together.
- Keep human checking and clear accountability wherever AI-assisted work enters a public record or affects a citizen service.
What we still do not know
The official record established the trainer cohort, but not the full multiplication and outcome trail.
- How many officers each Change Ambassador subsequently trained and how that coverage differed between ministries and agencies.
- How many of the 445,000 eligible officers became recurring users on approved tasks after completing appropriate learning.
- Which public services improved after checking and rework were counted, and what error, incident or complaint rates accompanied those gains.
Sources
- 1.Malaysian Government Upskills 200 Change Ambassadors In Generative AI Tools To Drive Public Service Transformation Ministry of Digital Malaysia, 12 March 2025
- 2.AI At Work 2.0 Train-the-Trainer equipped 200 Change Ambassadors Gobind Singh Deo, 13 March 2025
- 3.445,000 Public Officers in Malaysia to Benefit from Generative AI Under the AI at Work 2.0 Initiative Google Cloud, 5 February 2025
