GUIDE · PUBLIC SECTOR

Claude for Malaysian Government Agencies: When It Fits and How to Start

Updated 28 October 2026

Checked by Hanny Ramzy on 26 October 2026. Drafted with AI help from Anthropic's pricing, privacy and commercial terms pages, SPA Bil. 2/2021 v2.0 and the JDN GPAISA guideline.

Agencies now send us requests for quotation for Claude Team, usually for an IT unit or a small group of officers. This guide is for the officer who must explain the choice to management. It shows what Claude does well, a short test in Bahasa Melayu, how to give Claude your letter format, the data rules, the seat types and a plan for a small pilot.

Claude is not the correct tool for every agency. Where Gemini or Copilot fits better, this guide says so.

What Claude does well in agency work

Claude is an AI assistant from Anthropic. These are the tasks where it is strongest for an agency:

  • Long documents. Claude can read a full tender document, a long report or a set of circulars in one chat. The context window is up to 1 million tokens, which varies by model. That is hundreds of pages.
  • Careful writing. It drafts letters, minutes, briefing notes and replies to the public. It follows a template when you give one.
  • Data in Excel and PDF. It reads a spreadsheet or a PDF table, explains the numbers and makes charts.
  • Projects. A unit can keep its SOPs, templates and sample letters in one Project. Each officer in the unit then starts from the same material.
  • Connectors. On the Team plan, Claude can search Google Drive, Microsoft 365, Gmail and other services. It sees only the files that the officer can open. An admin controls which connectors are on.
  • Claude Code. Each Team seat includes Claude Code, so an IT unit can use it to write and fix scripts.

Claude also works inside Google Docs, Sheets and Slides. Our guide to Claude in Google Docs shows how.

Our test: three tasks in Bahasa Melayu

On 26 October 2026, we gave Claude (model Opus 5.5) three tasks in Malay. We ran them in Claude Code, in an empty folder, with no instructions and no files. We used sample details, not real agency data. The panels below show our prompts and Claude’s answers as text. We shortened only the long letter in Task 1.

Task 1: a surat rasmi. We asked for a letter that invites five agencies to a meeting on data sharing.

Claude Opus 5.5Task 1 · Surat rasmi
Tulis surat rasmi kerajaan untuk menjemput lima agensi ke Mesyuarat Perkongsian Data Antara Agensi Bil. 1/2026. Jangan reka nombor rujukan, tarikh atau tempat.

Ruj. Kami: [No. Rujukan Fail]Tarikh: [Tarikh Surat]

SEPERTI SENARAI EDARAN

YBhg. Dato' / Datin / Tuan / Puan,

JEMPUTAN KE MESYUARAT PERKONGSIAN DATA ANTARA AGENSI BIL. 1/2026

Dengan hormatnya saya merujuk kepada perkara di atas.

2.Sukacita dimaklumkan bahawa [Nama Kementerian / Jabatan Penganjur] akan mengadakan Mesyuarat Perkongsian Data Antara Agensi Bil. 1/2026. Butiran mesyuarat adalah seperti yang berikut:

Tarikh: [Hari, Tarikh Mesyuarat]Masa: [Masa Mula – Masa Tamat]Tempat: [Nama Bilik Mesyuarat / Lokasi / Pautan Dalam Talian]Pengerusi: [Nama dan Jawatan Pengerusi]

Paragraphs 3 to 6 are not shown: the purpose, the invitation, the confirmation of attendance and the thanks.

Sekian.

"[SLOGAN RASMI KERAJAAN SEMASA]"
"BERKHIDMAT UNTUK NEGARA"

Saya yang menjalankan amanah,

([NAMA PENANDATANGAN])
[Jawatan]

Not shown: a distribution list for the 5 agencies and an agenda (Lampiran A).

Sebelum surat ini dimuktamadkan, sila semak perkara berikut: … Slogan: Gunakan slogan rasmi yang ditetapkan oleh kementerian anda sekarang.

Our test, 26 October 2026. Sample details only. Run in Claude Code.
Claude used the government letter format and left a blank field for each detail that we did not give. We shortened the letter.

The structure was correct: “Ruj. Kami”, the title in capitals, numbered paragraphs from 2, “BERKHIDMAT UNTUK NEGARA” and “Saya yang menjalankan amanah”. Claude did not invent a file reference, a date or a venue. It also did not fill in the government slogan. It left a blank field and told us to use the slogan that the ministry sets. This is the reason to give Claude your agency’s format first (see the next section).

Task 2: a summary of a circular. We gave Claude 8 sebut harga rules from PK 2.9, in English, and asked for five points in Malay for a briefing.

Claude Opus 5.5Task 2 · Summary of a circular
Ringkaskan peraturan sebut harga PK 2.9 di bawah dalam lima perkara untuk taklimat.Below the prompt: 8 rules from PK 2.9, in English.

Ringkasan Peraturan Sebut Harga (PK 2.9)

  1. Had nilai dan kelayakan pembekal
    • RM50,000 hingga RM100,000: Hanya pembekal Bumiputera tempatan yang berdaftar dengan MOF di negeri berkenaan boleh menyertai. Jika pembekal tidak mencukupi, agensi boleh mempelawa pembekal dari negeri berhampiran atau negeri bersempadan (para 1.4).
    • Lebih RM100,000 hingga RM500,000: Pembekal tempatan berdaftar MOF boleh menyertai. Taraf Bumiputera tidak disyaratkan.
  2. Pendaftaran dan kod bidang
    • Pembekal mesti mempunyai kod bidang MOF yang berkaitan dengan barangan atau perkhidmatan yang diperolehi (para 2.1(iii)).
  3. Proses dalam talian dan tempoh notis
    • Keseluruhan proses sebut harga hendaklah dibuat dalam talian melalui ePerolehan (para 2.1(ii)).
    • Notis sebut harga mesti dikeluarkan sekurang-kurangnya 7 hari.
  4. Penilaian dan sah laku tawaran
    • Jawatankuasa Sebut Harga menilai tawaran sama ada dalam satu atau dua peringkat.
    • Tawaran sah selama maksimum 90 hari. Agensi tidak boleh melanjutkan tempoh ini (para 8.1 dan 8.2).
  5. Bon pelaksanaan
    • RM200,000 hingga RM500,000: 2.5% daripada nilai kontrak.
    • Melebihi RM500,000: 5% daripada nilai kontrak.

Teks asal menyebut kadar 5% untuk nilai melebihi RM500,000, tetapi had atas sebut harga ialah RM500,000. Jadi kadar itu nampaknya terpakai di luar sebut harga, iaitu untuk tender. Anda mungkin mahu menyemaknya dengan dokumen PK 2.9 sebelum taklimat.

Our test, 26 October 2026. Sample details only. Run in Claude Code.
The summary kept each value and paragraph number, and it found a problem in our text.

Claude kept all the values and the paragraph numbers. It also found a problem in our own text. We wrote a 5% bond for contracts above RM500,000, but a sebut harga stops at RM500,000. Claude said that the 5% rate probably applies to a tender, and told us to check PK 2.9 before the briefing.

This is the Claude Code window for Task 2. The panel above shows the same answer:

Claude Code terminal with the Task 2 prompt and Claude's answer in Malay
The real run in Claude Code, Opus 5.5, 26 October 2026.

Task 3: a reply to a public complaint. We asked for a polite reply about a late payment, with no promised date.

Claude Opus 5.5Task 3 · Reply to a complaint
Tulis balasan sopan kepada aduan awam tentang bayaran yang lewat. Jangan janjikan tarikh.

Subjek: Maklum Balas Aduan Berkenaan Kelewatan Bayaran

Tuan/Puan [Nama],

Terima kasih kerana menghubungi kami dan memaklumkan perkara ini. Kami memohon maaf atas kelewatan bayaran yang tuan/puan alami serta kesulitan yang timbul akibatnya.

Aduan tuan/puan telah kami terima dan direkodkan dengan nombor rujukan [No. Rujukan]. Pihak kami sedang menyemak status bayaran tersebut bersama bahagian yang berkaitan untuk mengenal pasti punca kelewatan dan mempercepatkan penyelesaiannya.

Kami akan memaklumkan perkembangan terkini kepada tuan/puan sebaik sahaja semakan selesai. Sementara itu, jika tuan/puan mempunyai sebarang maklumat tambahan seperti nombor invois, tarikh tuntutan atau dokumen sokongan, sila kemukakan kepada kami melalui [e-mel / nombor telefon] dengan menyertakan nombor rujukan di atas. Maklumat ini akan membantu kami menyelesaikan semakan dengan lebih cepat.

Kami menghargai kesabaran dan kerjasama tuan/puan. Maklum balas tuan/puan amat penting untuk kami menambah baik perkhidmatan.

Sekian, terima kasih.

Yang benar,
[Nama Pegawai]
[Jawatan]
[Nama Jabatan / Organisasi]
[No. Telefon] | [E-mel]

Surat ini tidak menyebut tarikh bayaran atau tempoh masa. Ia hanya berjanji untuk memaklumkan perkembangan selepas semakan selesai. Gantikan medan dalam [kurungan] sebelum dihantar.

Our test, 26 October 2026. Sample details only. Run in Claude Code.
The reply said sorry, gave the next step and a contact point, and did not promise a date.

What we saw:

  • The Malay was formal and correct, in the style of a government office.
  • Claude did not invent file numbers, names, dates or the slogan. It left blank fields and told us what to check.
  • An officer must still check each fact, name, date and reference. Claude can make mistakes, and it does not know your agency’s records unless you give them.

Give Claude your agency’s format first

Our test used no context, so Claude wrote a good general letter, not your agency’s letter. Before officers use Claude for surat rasmi, give it the correct material. Do these steps:

  1. Get the surat rasmi guideline that your agency follows. For a federal agency, this is the JPA Panduan Pentadbiran dan Pengurusan Pejabat (PPSM PO.1.1.2, Bahagian VI, 2022). Add your agency’s own rules, if it has them.
  2. Select 2 or 3 approved letters from your files. Remove personal data first.
  3. Create a Project in Claude, for example “Surat rasmi”. Upload the guideline and the sample letters to the Project.
  4. Write the Project instructions. Include the current slogan, the letterhead text, the file reference format and the title of the officer who signs.
  5. Share the Project with the officers in the unit. On the Team plan, you can also share a Project with all users in the organisation.
Cover of the JPA Pekeliling Perkhidmatan Sumber Manusia: Panduan Pentadbiran dan Pengurusan Pejabat, Versi 1.0 (2022)
The JPA guideline for federal agencies. Bahagian VI gives the surat rasmi format. Open it on the JPA site.

A skill can also help with the language. A skill is a set of instructions that Claude uses for one type of task. For example, the community skill dbp-translator translates English to Malay with DBP spelling. It uses terms from Kamus Dewan and PRPM. It is not an official DBP product, and it does not cover the letter format.

The dbp-translator-claude repository on GitHub: a Claude skill by Chuah Kee Man that translates English to Bahasa Melayu with DBP conventions, MIT licence
The community skill on GitHub, 26 October 2026. Read the SKILL.md file before you use it.

On the Team plan, the owner can publish a skill for all users. Skills need the code execution setting to be on. Ask your IT unit to read a skill before the owner publishes it.

Data and security: what the Team plan gives

These points are from Anthropic’s own pages on 26 October 2026:

  • No training by default. Anthropic does not train its models on inputs and outputs from its commercial plans. If a user sends feedback, Anthropic can use that chat. An owner can turn off feedback with the “Rate chats” setting.
  • You own the outputs. The commercial terms say that the customer keeps its inputs and owns its outputs.
  • Deletion. A deleted chat leaves the chat history at once, and Anthropic deletes it from its storage within 30 days.
  • Sign-in control. The Team plan has single sign-on (SSO) and domain verification. Admins can set spending limits for each user.
  • Certification. Anthropic has ISO/IEC 42001:2023 certification for the management of AI systems (January 2025).

What the Team plan does not give

Read these limits before you write your paper:

  • No Malaysian data location. The Team plan does not let you select a country for your data. Anthropic says that it can process data in other countries.
  • No audit logs and no SCIM. These are only on the Enterprise plan, which needs minimum 20 seats.
  • Not for classified data. SPA Bil. 2/2021 does not permit rahsia rasmi on a public cloud. Use Claude only for open and internal work: drafts, summaries of public documents, meeting notes and translations.

The JDN GPAISA guideline (February 2025) also asks agencies to check privacy and security before they use AI. Our procurement guide for AI licences gives the contract terms to ask for.

Claude, Gemini or Copilot: which one fits?

The best tool depends on the system that the agency uses now. This table gives a fair first choice:

Your situationA good first choice
The agency uses Google Workspace, and all staff want AI in Gmail, Docs and MeetGemini, which comes with Business Standard and higher
The agency uses Microsoft 365, and all staff want AI in Word, Excel and OutlookMicrosoft 365 Copilot
A unit works with long documents and writes many letters and reportsClaude Team, Standard seat
An IT unit writes scripts and code, or some officers use AI all dayClaude Team, Premium seat

Many organisations use two tools. They give Gemini or Copilot to all staff, and Claude to a small group of heavy users. Our Gemini guide shows what Gemini does in Google Workspace.

Standard seat or Premium seat?

The Team plan is for 2 to 150 seats. You can mix the two seat types, and you can add seats during the term. These are the prices on claude.com on 26 October 2026:

SeatBilled monthlyBilled annuallyUsage
StandardUSD 25 per seat per monthUSD 20 per seat per month (USD 240 a year)More than Claude Pro
PremiumUSD 125 per seat per monthUSD 100 per seat per month (USD 1,200 a year)5 times the Standard seat

The prices do not include tax, and Anthropic can change them. Usage limits reset every five hours, with a weekly limit also. Give the Standard seat to most officers. Give the Premium seat to the IT unit and to officers who reach the limit often.

Start with a small pilot

A pilot gives management numbers, not opinions. Do these steps in sequence:

  1. Select 5 to 10 officers in one unit.
  2. Select 3 tasks that the unit does each week, for example letters, minutes and report summaries.
  3. Record the time that each task takes now.
  4. Write a usage rule. Officers must use work accounts only. Officers must not put classified data in Claude.
  5. Run the pilot for 3 months.
  6. Record the time again, and compare the results.
  7. Use the results to decide the number of seats for the next year.

Agree the licence term with your finance unit. Budget that the agency does not spend lapses on 31 December, so a term inside one financial year is the simplest.

How to buy

Ten Standard seats for one year cost USD 2,400 at the vendor price. That is usually inside the direct purchase limit of RM50,000 a year. For larger orders, our procurement guide shows the method for each value.

Web Impian Sdn. Bhd. is MOF-registered (357-0002304403) with Bumiputera status, and we have the codes 210103 and 210104 for software. On our quotation, the Claude licence is at Anthropic’s price. Our support and management work is a separate line, so your committee can see the two amounts. We accept LO and PO. We are working toward a partnership with Anthropic. Our AI licences for the public sector page has the details, and Claude for organisations shows the plans and the setup that we do.

How we wrote this guide

We drafted this guide with AI help, and we used these sources on 26 October 2026:

The Malay test used Claude Opus 5.5 in Claude Code, with no instructions, no files and sample details. Vendors change prices and features, so check the vendor page on the day that you write your paper.

Next steps

If you know the number of officers, WhatsApp us. We will suggest the mix of Standard and Premium seats and send a quotation in Ringgit with the MOF documents.

Hanny Ramzy

Hanny Ramzy runs Office Niaga at Web Impian Sdn. Bhd. (SSM 920163-H) in Kota Bharu, Malaysia, helping businesses and agencies move to Google Workspace and business email with Ringgit billing and local support.

Straight answers

Frequently asked questions

1Does Anthropic train its models on our chats?

Not by default. Anthropic does not train its models on the inputs and outputs of its commercial plans, which include Team. If a user sends feedback, for example a thumbs up or down, Anthropic can use that chat. An owner can turn off the "Rate chats" option for the organisation.

2Can officers put official secrets in Claude?

No. Claude is a public cloud service, and the Team plan has no data location in Malaysia. SPA Bil. 2/2021 does not permit rahsia rasmi on a public cloud. Classify the data first. Use Claude only for open and internal work.

3Does Claude understand Bahasa Melayu?

Yes. In our test, Claude wrote a surat rasmi, a summary and a reply to a complaint in formal Malay. For your agency's own format, put your guideline and approved sample letters in a Claude Project first. The officer must still check the facts, the names and the references before the letter goes out.

4Is the Team plan sufficient, or do we need Enterprise?

The Team plan is for 2 to 150 users, and it has SSO and admin controls. It does not have audit logs or SCIM. Enterprise adds these, but it needs minimum 20 seats and it charges usage at API rates. Most pilots start on Team.

5Can officers use a personal Claude Pro account for office work?

No. A personal account follows Anthropic's consumer terms, not a business contract. The agency cannot control the account or remove the data. Give each officer a seat in the agency Team plan.

Plan a Claude pilot for your agency

Tell us the number of officers and the tasks. We send a quotation in Ringgit with the vendor price, our support line and the MOF documents.

Claude
Request quote →