Public Sector / AI news for Malaysia
From the archive · Event date 5 February 2025
Malaysia opened Google AI tools to 445,000 civil servants. Access was only the start
AI at Work 2.0 made Google Workspace with Gemini and training available across the public service. The scale was significant, but access alone did not show that services became faster, fairer or more reliable.
In brief
- The Ministry of Digital, NAIO and Google Cloud launched AI at Work 2.0 on 5 February 2025, making Google Workspace with Gemini tools and training accessible to up to 445,000 public officers.[1][2]
- The programme covered drafting, summarising, data analysis, meetings and research support, with dedicated workshops intended to help officers identify useful tasks and write effective prompts.[1][2]
- The headline number described possible access, not 445,000 trained active users or independently verified improvements in Malaysian public services.[1][2]
AI at Work 2.0 created large-scale access, with training meant to turn tools into useful work
Malaysia's Ministry of Digital, the National AI Office and Google Cloud launched AI at Work 2.0 in Kuala Lumpur on 5 February 2025. The programme expanded an earlier pilot and made a wider Google Workspace with Gemini toolset available to as many as 445,000 public officers.[1][2][3]
The planned uses were ordinary but consequential: drafting policy papers and communications, analysing information, summarising documents, supporting meetings, producing explanatory material and extracting insights from approved sources. Google also said the rollout would include dedicated training and consultation rather than account access alone.[1][2]
The important distinction is between reach and result. Up to 445,000 officers could receive access, but the launch material did not establish how many became active users, which agencies changed a service, how outputs were checked, or whether citizens experienced faster or better decisions. Those are the measures needed before a technology rollout becomes a public-service success story.[1][2]
The rollout covered familiar office work, not autonomous government decisions
Google described Gemini assistance inside Gmail, Docs, Sheets, Slides, Drive, Meet and Chat. The stated functions included drafting, summarising, analysing information and helping with meeting notes. NotebookLM Plus was presented as a way to work across selected reference material, while Google Vids and the Gemini app extended the toolset to communications, data extraction and idea development.[2]
These are assistant functions. The official material did not announce an autonomous system deciding licence applications, benefits, enforcement, procurement or individual eligibility. A generated draft still requires an officer to verify the source, context, policy rule and final wording before it becomes an official act.[1][2]
That boundary matters because government documents can affect rights, money and access to services. A useful deployment therefore needs a clear list of permitted tasks, prohibited high-risk uses, required human approvals, records of source material and a route for correcting an inaccurate output. The launch announced tools and training; it did not publish that complete operating control set.[1][2]
The early pilot numbers were encouraging, but they were supplier-reported and small
Google Cloud said the earlier AI at Work pilot involved 270 public officers from several agencies, including the National Digital Department. Its release reported that 97% experienced time savings averaging 3.25 hours per person each week and that 91% said AI improved their work quality.[2]
Those figures describe feedback and usage analytics from a limited pilot, not an independent evaluation of all 445,000 potential users. The release did not provide the questionnaire, task mix, comparison group, distribution of time saved, error rate or service-level outcome. It is therefore reasonable to treat the numbers as an early signal, not a nationwide performance result.[2]
Scaling also changes the risk profile. A tool that helps a motivated pilot group draft routine material may behave differently when used across hundreds of agencies, job types and data environments. Training completion, active use, correction rates and supervisor review should be reported separately so that a large eligibility number does not hide uneven adoption.[1][2]
Malaysia can prove value by measuring one public-service workflow at a time
A credible scorecard starts with a specific service. An agency could compare the time to prepare a briefing, the number of factual corrections, approval turnaround, staff satisfaction and citizen response before and after an AI-assisted workflow. Measures should include both benefit and harm: time saved, rework, hallucinations, privacy incidents and complaints.[1][2]
The Ministry said MyGovUC spans 25 ministries, three state secretary offices and 200 government agencies. That reach makes consistent governance important, but it also creates an opportunity for comparable evidence. Agencies can publish the same small set of measures while explaining where their workflows differ.[1]
The most useful public update would therefore go beyond licences issued or officers invited. It would show active trained users, approved use cases, hours saved after verification, material error rates, human-review compliance and changes in service delivery. That would let Malaysians see whether AI improved government work without confusing access with impact.[1][2]
Why Malaysia should care
This was one of Malaysia's largest public-sector AI access programmes. Its value should be judged by better public service, not the number of accounts enabled. Agencies need task-level evidence, human review, secure information handling and published before-and-after measures that citizens can understand.
Public officers
The programme could reduce routine drafting and research work, but officers remain responsible for source checking and official decisions.[1][2]
Practical move: Use AI only for approved tasks, preserve source links and record every material correction before submission.
Agency leaders
Account access is a starting condition, not proof that productivity or citizen service improved.[1][2]
Practical move: Choose bounded workflows and publish before-and-after time, quality, error and review measures.
Malaysians using public services
The rollout may improve response preparation, but citizens need transparency when AI materially shapes a service interaction.[1][2]
Practical move: Look for clear correction and appeal routes, especially where an AI-assisted output affects eligibility, rights or payments.
What Malaysians can do now
- Separate officers who can access the tools from those who completed training and use them actively on approved tasks.
- Measure time saved only after verification and report factual corrections, rework, incidents and complaints alongside productivity.
- Keep human sign-off and a clear correction route for any AI-assisted communication or decision that affects the public.
What we still do not know
The launch established scale and intended uses, but not nationwide service outcomes.
- How many of the 445,000 eligible officers became trained monthly active users across each ministry and agency.
- Which public-service workflows produced verified improvements after accounting for checking, correction and rework.
- What material error, privacy-incident, human-review and citizen-complaint rates were observed during the wider rollout.
Sources
- 1.Ministry Of Digital, Through NAIO, Secures AI Training For 445,000 Civil Servants Under Google AI At Work 2.0 Initiative Ministry of Digital Malaysia, 5 February 2025
- 2.445,000 Public Officers in Malaysia to Benefit from Generative AI Under the AI at Work 2.0 Initiative Google Cloud, 5 February 2025
- 3.Launch of AI at Work for Public Services 2.0 AI Malaysia / National AI Office, 5 February 2025
