Public Sector AI17 min read

AI in the Public Sector in Kenya: A Practical Guide for Government Officers (2026)

Young people in a group mentorship discussion
On this page

TL;DR:

  • Kenya's public sector can use AI today for low-risk paperwork: drafting memos and letters, summarising long reports and bills, tidying minutes, analysing budget spreadsheets and preparing public notices in English and Kiswahili.
  • The rules are still forming. The National AI Strategy 2025–2030 is in place, the AI policy and the ODPC's AI guidance are drafts, and the AI Bill 2026 is a proposal, not law. The Data Protection Act already applies in full.
  • Citizens' personal data, confidential files and anything classified should never go into a public AI tool.
  • A human officer reads, checks and signs off every output. AI drafts; the officer is accountable.
  • Start small: one written AI use policy, a few trained champions, one pilot task, and an honest record of time saved.

Who this guide is for

This guide is for people who keep government running day to day. That means officers in ministries and state departments, state corporations and agencies, county governments, constitutional commissions and independent offices. It also covers county assembly staff: clerks, committee staff, researchers and Hansard teams.

You do not need any technical background. If you write letters, read long reports, prepare budgets or answer the public, this is for you.

Where Kenya stands on AI right now (October 2026)

It helps to know which documents are in force and which are still proposals. News coverage often blurs the difference.

Document What it is Status as of late October 2026
Kenya National AI Strategy 2025–2030 The government's plan for AI infrastructure, data, research and skills Launched on 27 March 2025
Kenya Artificial Intelligence and Other Emerging Technologies Policy, 2026 National policy direction on how AI is governed, including citizen-data safeguards Draft; public consultation ran from late July to August 2026
The Artificial Intelligence Bill, 2026 (Senate Bill No. 4) A proposed law creating an AI Commissioner and rules for high-risk systems A bill, not law; in mid-October 2026 the Senate ICT committee recommended that the Senate not proceed with it
ODPC Guidance Note on Artificial Intelligence Guidance from the Data Commissioner on applying data protection law to AI Published in draft for consultation; comments closed 17 August 2026
Data Protection Act (Cap. 411C) The law on how personal data is collected and used In force; applies to every public body today

The National AI Strategy 2025–2030

The Ministry of Information, Communications and the Digital Economy launched the strategy at KICC in Nairobi on 27 March 2025. ICT Cabinet Secretary William Kabogo led the launch. Coverage at the time reported that at least KES 152 billion would be needed to implement it by 2030. The strategy rests on three pillars: AI digital infrastructure, data, and AI research and innovation. Talent development is one of four cross-cutting enablers, alongside governance, investment, and ethics, equity and inclusion.

For a public officer, the strategy is the "why". It signals that government expects AI to be used in service delivery, and that skills are part of the plan.

The draft AI and Emerging Technologies Policy, 2026

In July 2026 the Ministry released the draft Kenya Artificial Intelligence and Other Emerging Technologies Policy, 2026, and invited public comment until early August 2026.

Reports on the draft highlight points that matter to anyone handling citizens' data:

  • Organisations that develop or deploy AI must comply with the Data Protection Act.
  • AI platform providers must ensure that data generated in Kenya is governed under Kenyan law.
  • Sensitive datasets may face data localisation requirements, with investment planned in local data centres and cloud capacity.
  • People should be told when they are dealing with an AI system and not a human, and should be able to seek review of important algorithmic decisions.
  • A Kenya AI Safety Institute is proposed to assess high-risk systems before wide use.
  • Public administration is named as a priority sector for AI.

The draft does not itself create binding legal obligations. It sets direction for laws and regulations that would follow.

The AI Bill, 2026

Senate Bill No. 4 of 2026, sponsored by Senator Karen Nyamu, was published on 19 February 2026 and read a first time on 2 April 2026. It is not law. See the AI Bill section further down for what it proposed and where it stands.

The ODPC's draft Guidance Note on AI

The Office of the Data Protection Commissioner (ODPC) published a draft Guidance Note on Artificial Intelligence, dated July 2026, and took comments until 17 August 2026. For government, the key point is that public bodies may rely on "public interest or exercise of official authority" as the lawful basis for AI processing their functions require, but must still meet every other duty, including transparency and Data Protection Impact Assessments (DPIAs).

The note also has a section on generative AI (tools such as ChatGPT, Gemini, Claude and Copilot) that asks organisations to guard against hallucinations, data leakage and inaccurate outputs about real people.

The training push inside government

Government is investing in AI skills for its own officers. If your department has places on these public programmes, take them.

  • Regional Centre of Competence (RCOC) for Digital and AI Skilling. Hosted by the Kenya School of Government (KSG), the centre says it targets training 100,000 public servants over the next three years, using a mix of virtual and in-person sessions. It recommends AI skilling for officers in all cadres, in ministries, state departments and county governments.
  • DigiKen course on AI and digital transformation. Developed by UNESCO and the University of Oxford and delivered with KSG. The first cohort had 2,730 learners, and 1,239 graduated at KSG on 8 October 2026. A second round began on 21 October 2026, designed to reach an additional 9,000 public servants at national and county levels.
  • KSG and Huawei Kenya. On 19 May 2026 the two signed an MoU covering training and certification of 2,000 government officials and 50 KSG technical staff between 2026 and 2029, with AI and cybersecurity among the focus areas.

Courses build the foundation. Everyday habits come from practice on your own department's real work.

What AI is useful for in a public office

The best starting point is repetitive writing and reading work that is low risk and easy to check. Here are common tasks and how AI helps.

Task How AI helps What the officer still does
Memos and letters Produces a first draft in the right tone and format Checks facts, references and the signatory's position
Minutes of meetings Turns rough notes into structured minutes with action points Confirms decisions and owners with the chair
Long reports and bills Summarises the main points, sections and implications Reads the sections that matter before advising
Budget analysis in Excel Explains formulas, builds variance tables, spots unusual lines Verifies every figure against the source
Public communication Drafts notices, press statements and social posts in English and Kiswahili Approves wording and checks the Kiswahili with a fluent speaker
Citizen FAQs Turns a policy or service charter into plain questions and answers Confirms every answer is correct and current
Procurement documents Checks a tender document against a checklist you provide, flags missing items Makes every procurement judgement; AI never evaluates bids
M&E reporting Organises indicator data and drafts the narrative Checks numbers and conclusions against the evidence

In every row, AI speeds up the first draft and the officer remains the decision-maker.

How to write a prompt that works: the four Cs

Vague requests get vague answers. At FuKazee we teach a simple method, the four Cs:

  • Clarity: say exactly what you want produced.
  • Context: give the background the tool cannot guess (who you are, who it is for, why).
  • Constraints: set limits on length, tone, format and what to leave out.
  • Customisation: shape it for the reader and the place it will be used.

A useful question for format: where will this answer be used, and who will read it? A memo for a Principal Secretary reads differently from a notice on a county noticeboard.

Five worked prompts for public officers

Replace the square-bracket placeholders with your own non-sensitive details. Never paste citizens' names, ID numbers or confidential file contents.

1. Drafting an internal memo

I am a Senior Administrative Officer in a county Department of Health. Draft an internal memo from the Chief Officer to all sub-county health administrators. The purpose is to remind them that quarterly expenditure returns are due on [date]. Context: last quarter, several returns arrived late, which delayed the county's consolidated report. Constraints: no more than 250 words, formal tone, use the standard memo heading (To, From, Date, Ref, Subject), end with a clear deadline and a contact person [title only]. Customisation: readers are busy administrators who will skim, so put the deadline in the first paragraph.

2. Turning rough notes into minutes

Below are my rough notes from a departmental management meeting. Turn them into formal minutes. Context: this is a state department; minutes go to the Principal Secretary's office. Constraints: use numbered minute items (MIN 01/2026 and so on), record decisions and action points only, do not invent any decision or name that is not in my notes, and mark anything unclear with [CHECK]. Customisation: add a table at the end listing each action, the responsible office and the deadline. Notes: [paste notes with personal details removed]

"Do not invent" and the [CHECK] marker make gaps visible, so you can confirm them with the chair.

3. Summarising a bill or long report

I will paste the text of a publicly available bill. Summarise it for a county assembly committee that must give a view within a week. Context: committee members are not lawyers. Constraints: maximum one page; list (1) the purpose of the bill, (2) the main provisions with their clause numbers, (3) what it would mean for county governments, and (4) questions the committee should ask. Quote clause numbers exactly and say "not stated in the text" where the bill is silent. Customisation: plain English, short bullet points.

Use public documents only. Then read the clauses the summary points to before you rely on it.

4. Analysing a budget in Excel

I have an Excel sheet with columns: Programme, Approved Budget (KES), Actual Expenditure (KES) for a county department. Context: I need to prepare a mid-year absorption report. Constraints: give me the exact Excel formulas to calculate absorption rate (actual divided by approved, as a percentage) and variance in KES for each row, and a formula to flag any programme below 40% absorption. Explain each formula in one sentence. Customisation: I use Microsoft Excel and I am comfortable with basic formulas but not pivot tables.

Here AI supplies the method while the figures stay in your own spreadsheet.

5. A public notice in English and Kiswahili

Draft a short public notice from [County Government of X], Department of Water, informing residents of [ward] that water supply will be interrupted on [date] from [time] to [time] for pipe repairs. Context: notice will go on the county website, Facebook page and printed on noticeboards. Constraints: under 120 words, calm and respectful tone, include an alternative water point [location] and an enquiries line [official number]. Customisation: give an English version and a Kiswahili version that reads naturally for residents, not a word-for-word translation.

A fluent colleague should always read the Kiswahili version before it goes out.

Which tools to use, and the free versions

You do not need to buy anything to start learning. No single tool is the right one; use what your institution approves.

  • Microsoft 365 Copilot Chat. If your office uses Microsoft 365 with work accounts, Microsoft says Copilot Chat is included at no extra cost when you sign in with your work account. The free version draws on the web and on files you upload. Reaching your organisation's emails and documents needs a paid Microsoft 365 Copilot licence. Your ICT team may have switched features on or off.
  • ChatGPT, Gemini and Claude. Each has a free tier that handles drafting, summarising and explaining well.

Before using any tool for work, check two things with your ICT office: whether the tool is approved, and what happens to what you type. Some free consumer tools may use conversations to improve their models unless that setting is turned off. That is one reason sensitive material must stay out of them.

What must never go into a public AI tool

As a simple rule, if you would not post it on a public noticeboard, do not paste it into a public AI tool. That includes:

  • Citizens' personal data: names with ID numbers, KRA PINs, phone numbers, addresses, health records, case files, land records, payroll details.
  • Sensitive personal data under the Data Protection Act, such as health, ethnic origin and biometric data.
  • Classified or confidential government documents, Cabinet papers, draft policy not yet cleared for release, and security information.
  • Procurement information before award, including bids, prices and evaluation scores.
  • Passwords, system credentials and internal network details.

When you need AI help with a document that contains personal data, remove or replace the identifying details first, or use a tool your institution has approved and assessed for that purpose.

Citizens' data and the Data Protection Act

Every ministry, county and agency that decides how personal data is used is a data controller under the Data Protection Act, and AI adds new ways to break those duties. The ODPC's draft guidance expects public bodies to carry out a DPIA for higher-risk AI uses, be open with citizens about AI processing, keep humans in the loop for decisions that affect people, and take care with personal data that leaves Kenya through foreign cloud services. Our companion guide explains what this means for an ordinary office, with a checklist for ministries and counties: Data protection and AI in Kenya's public sector.

Running open-source AI on your own computer or server

Some offices worry, rightly, about sending documents to tools hosted abroad. One option is an open-source AI model that runs on your own computer or office server, so the text never leaves the building. These models are smaller than the big online tools and need a reasonably powerful machine, but they can handle many summarising and drafting tasks. This route needs your ICT team's involvement and approval. Our step-by-step guide covers what hardware you need, which free tools to look at, and where the limits are: How to run open-source AI models locally.

Kenya's AI Bill 2026 in brief

The Artificial Intelligence Bill, 2026 proposed a dedicated AI regulator, risk-based rules for high-risk systems and penalties of up to KES 5 million. Critics argued it would add regulators alongside existing bodies such as the ODPC and the Communications Authority, creating overlapping duties and red tape, and that its penalties would discourage local innovation. In mid-October 2026, The Star and Techweez reported that the Senate's Standing Committee on ICT recommended that the Senate not proceed with it, finding it too broad, too early and too punitive. The full Senate still has to vote. A bill can be redrafted and return, so it is worth knowing what it proposed. Read our plain explainer: Kenya's AI Bill 2026 explained.

The risks, and how to manage them

Risk What it looks like What to do
Hallucinations AI states a wrong figure, an invented clause or a fake reference with full confidence Check every fact, figure and citation against the source document
Bias Drafts that assume things about gender, region or community, or weaker output in Kiswahili than in English Read with fresh eyes; have a colleague review public-facing text
Confidentiality Personal or classified data pasted into a public tool Follow the "never paste" list; use approved tools only
Over-reliance Officers stop reading the full report and trust the summary Use summaries to guide reading, not replace it, for anything you advise on
Records and accountability AI-drafted documents enter the official record without anyone owning them A named officer reviews and signs off; keep your normal records rules
Procurement of AI Buying an AI system without knowing where data goes or who is liable Ask about data location, security, DPIA support and exit terms before signing

Take special care with decisions about people. Section 35 of the Data Protection Act gives people the right not to be subject to decisions based solely on automated processing that significantly affect them. Eligibility, licensing, hiring and disciplinary decisions should always rest on a human officer's judgement.

A 30/60/90-day plan for a department

Days 1 to 30: set the ground rules

  • Write a one- or two-page AI use policy. Cover approved tools, the "never paste" list, the rule that a human signs off, and who to ask when unsure. Check first whether your ministry, county or ICT office has already issued guidance.
  • Name two or three AI champions. Pick curious, careful people from different units, not only ICT.
  • Choose one pilot task. Good first pilots are high-volume and low-risk: drafting routine letters, preparing minutes, or summarising public reports.
  • Record a baseline. Before using AI, note how long the pilot task takes today.

Days 31 to 60: train and pilot

  • Run hands-on training for the pilot team using real (non-sensitive) examples from your own work.
  • Build a shared prompt library: the five prompts above, adapted to your department.
  • Run the pilot. Champions hold a short weekly check-in to share what worked and what went wrong.
  • Log any errors AI made and how they were caught. That log is your best training material.

Days 61 to 90: review and decide

  • Compare time taken against your baseline.
  • Ask the pilot team and their supervisors about quality: were drafts better, worse or the same after review?
  • Update the policy with what you learned.
  • Decide whether to extend to a second task or more staff, and what training that needs.

Measuring time saved honestly

Claims about AI saving hours are easy to make and hard to trust. Measure your own.

  • Time the full task, including checking and correcting the AI's draft, not just the moment it produced text.
  • Compare like with like: the same type of letter or report, before and after.
  • Track quality, not only speed. A faster memo that needs two rounds of correction from the supervisor has not saved time.
  • Write down where AI did not help. That tells you where not to use it.

Modest results you can show make a stronger case to management than big claims nobody can verify.

Frequently asked questions

Is it legal for a Kenyan public officer to use ChatGPT or Copilot at work?

There is no Kenyan law that bans it. The Data Protection Act applies to any personal data you use, and your institution's ICT and confidentiality rules apply too. Use approved tools, keep personal and confidential data out, and check whether your ministry or county has its own guidance.

Is the AI Bill 2026 now law in Kenya?

No. It is Senate Bill No. 4 of 2026. It was published in February 2026 and read a first time in April. In mid-October 2026 the Senate ICT committee recommended that the Senate not proceed with it; the full Senate has not yet voted. Kenya currently has no dedicated AI law.

Can a county use AI to decide who gets a bursary or a licence?

Not on its own. Section 35 of the Data Protection Act gives people the right not to be subject to decisions based solely on automated processing that significantly affect them. AI can help organise information, but a human must make and own the decision, and applicants should be able to request human review.

Can AI write in Kiswahili well enough for public notices?

Often it produces a usable draft. Quality varies by tool and topic, so a fluent Kiswahili speaker should always review the text before it is published.

Where can public servants get AI training?

The Kenya School of Government runs public programmes, including the RCOC for Digital and AI Skilling and the DigiKen course. Departments also commission practical training built around their own daily tasks.

What is the first step for a department that wants to start?

Write a short AI use policy and pick one low-risk pilot task. Record how long that task takes now, train a small team, and measure honestly after 60 to 90 days.

Where FuKazee fits in

FuKazee trains public and private sector teams in practical AI for the work in front of them. We assume nobody in the room knows AI yet, we avoid jargon, and we show the free version of every tool so officers can keep practising after the training. We teach the four Cs, safe handling of data, and how to check every output. Since 2022 we have trained more than 3,000 people.

For ministries, state departments, state corporations, counties and county assemblies, we design sessions around your department's real tasks: memos, minutes, budgets, reports and public communication. See how it works on our page for AI training for government in Kenya.

To discuss training for your department, email us at lead@fukazee.com

Sources

Want guided support instead?

Not sure which programme fits? Tell us who you are and we will point you to the right path: corporate AI training, the ScholarShift Accelerator for postgraduates, or KCSE study plans and tuition.

Rodney MainaFounder & Director, FuKazee

Rodney has run FuKazee's training since 2022 and still teaches inside every programme line: corporate sessions, the postgraduate cohort and the KCSE examiner webinars. First Class Honours in Journalism and Mass Communication, University of Nairobi.