What Does Good AI Training in Uganda Look Like?

TL;DR:

• AI training Uganda employers can rely on starts with the data law, then moves to daily work: reports, minutes, customer replies, Excel and field notes. Free tools are enough to begin.
• The Data Protection and Privacy Act, 2019 is in force, and the Personal Data Protection Office (PDPO) says collecting personal data without registering is a criminal offence. The Act lists financial information as special personal data, so payroll files and mobile money statements stay out of public AI tools.
• As of 8 November 2026 we found no AI law in Uganda and no published national AI strategy. A draft was under review in May 2026, and a National Data Strategy launched in October 2026 as groundwork.
• Luganda is better served than most Ugandan languages and still needs a fluent reviewer. Several other languages are missing from the mainstream translation lists we checked.
• Design for phones, paid bundles and interruptions. Data averages about US$1.11 per GB, and January's election shutdown took the internet down for days, with WhatsApp blocked for some time after.

Who this guide is for

This guide is for the people who decide how staff in Uganda learn to use AI: HR and learning leads, heads of operations, compliance officers, data protection officers and country directors.

It covers employers in Kampala and in regional centres such as Gulu, Mbarara, Jinja, Mbale and Arua. The sectors are banking, telecoms and mobile money, agribusiness, manufacturing, NGOs and government-linked bodies.

Start here: is your organisation on the PDPO register?

Check this before you book any training. AI use sits on top of what your organisation has already told the regulator, and many organisations have told it nothing.

The PDPO sits under the National Information Technology Authority, Uganda (NITA-U). Section 29 of the Act (2019 numbering) says the register covers every person, institution or public body collecting or processing personal data, along with the purpose. Section 30 lets anyone inspect it.

Enforcement has become more visible:

• On 18 August 2026, PDPO director Baker Birikujja spoke at a chief executives' breakfast in Kampala. The Daily Monitor reported 14,697 registered controllers and processors out of about 50,000 expected. Eagle Online reported about 14,000 registered and more than 500,000 collecting data unregistered. Treat both gap figures as estimates.
• He called non-registration a criminal offence, and said the office works with the Criminal Investigations Directorate and the Director of Public Prosecutions.
• In July 2025 the PDPO announced its first criminal conviction under the law, of a director of a digital lender.
• On 18 July 2025, on a complaint by four Ugandan citizens, the PDPO held that the law reaches foreign companies handling Ugandans' data. It ordered Google to register within 30 days, according to DLA Piper's account of the decision.

Registration is modest in cost and effort. DLA Piper reported an initial fee of UGX 100,000 and annual renewal, and an earlier legal note describes a certificate valid for 12 months, renewed at least three months before expiry. Confirm the current steps on the PDPO portal, because these sources date from 2022 and 2025.

The undertaking most people forget

A 2022 legal note says applicants file a signed undertaking not to process or store personal data in another country unless it has adequate protection measures. That matters for AI. Most mainstream tools are run by foreign companies, so assume text you paste leaves Uganda unless the vendor tells you otherwise in writing.

Before training starts, four questions need answers:

1. Are we on the register, and is the certificate current?
2. Does the purpose on our entry still describe what we do with personal data?
3. Who is our data protection officer? Section 6 says the head of an institution must designate one.
4. Has anyone read the undertaking we signed against the AI tools staff already use?

What the 2019 Act means when staff open an AI tool

Six provisions change daily behaviour. Our section numbers follow the Act as assented in 2019. It now appears in the revised laws as Cap. 97, and we could not open that revised text, so check the numbers against your current copy.

Three details are easy to miss.

Financial information is on the special list. In a mobile money economy, that covers ordinary back-office material. Section 9 does allow processing under an employer's legal duties or with consent, and it gives non-profits some room for members and regular contacts. That ground excludes disclosure to a third party without consent, and a public AI tool with no processing agreement is likely to be one.

We could not confirm any official list of equivalent countries. DLA Piper's February 2026 summary said none had been published. In its July 2025 decision, as DLA Piper reports it, the PDPO said transfers do not need approval one by one, and it expects records of the legal basis, safeguards and justification for each transfer.

Section 27 has exceptions, including decisions made in connection with a contract or authorised by law. A lender may fall within them. Even so, an unchecked AI score is a poor way to decide someone's loan or job, and section 33 lets a person who suffers damage or distress go to court for compensation.

Under section 35, unlawfully obtaining or disclosing personal data carries a fine of up to 240 currency points (UGX 4.8 million), imprisonment for ten years, or both. Section 38 allows an additional fine of up to 2% of a convicted corporation's annual gross turnover.

This section is general information. Take advice from a Ugandan advocate before you set policy.

The Ugandan paste test

Uganda's AI rules: what exists and what is still being written

Uganda has data, cyber and sector rules that apply to AI use today. A dedicated AI framework is still in draft.

The timeline helps you read the headlines:

• February 2026: a national workshop on UNESCO's AI Readiness Assessment met in Kampala while the strategy was being drafted.
• April 2026: Permanent Secretary Dr Aminah Zawedde said it would be ready by June, with priorities in agriculture, health and public administration.
• May 2026: a three-day stakeholder workshop reviewed a draft.
• October 2026: the National Data Strategy launched, and iAfrica described the AI strategy as still forthcoming.

We also found no AI-specific guidance from the PDPO. Absence from our search does not prove none exists, so ask the office directly.

Nothing here stops you from starting. Your internal policy carries most of the weight for now, and a two-page document will do. Our guide to writing an AI usage policy was written for Kenya, and its structure transfers. Replace the law sections with the table above. Our piece on why AI rollouts fail without governance explains why this comes first.

What teams can actually use AI for

Start with written and spreadsheet work that a person can check against a source. Here is what that looks like in each sector.

Banks and financial institutions

Good first uses: plain-English summaries of public Bank of Uganda material, staff briefings, policy first drafts, Excel formulas for branch reports and role-play for customer conversations.

Stays with people: customer data, credit decisions and regulatory interpretation.

The Ugandan detail: the Governor said in a 2026 speech that police recorded 412 cybercrime cases in 2025, down from 474 in 2024 and up from 245 in 2023. AI literacy for bank staff should include spotting AI-written fraud, as well as writing faster. We did not review banking secrecy law, so ask compliance. Our page on AI training for banks is about Kenya, and the course shape is similar.

Telecoms, mobile money and agent networks

Good first uses: customer FAQ drafts, agent training notes, tariff explanations, SMS rewrites and sorting anonymised complaint themes.

Stays with people: customer records, final tariff wording and regulator correspondence.

The Ugandan detail: the Uganda Communications Commission (UCC) counted 2.37 billion mobile money transactions and 19.1 billion USSD sessions in the first quarter of 2026. Much of the customer journey happens on a tiny screen, so short, exact wording is a real skill. Front-line staff also explain charges constantly, including the 0.5% excise on withdrawals, which stayed in place after a proposal to halve it was dropped for 2026/27.

Agribusiness and coffee

Good first uses: field officer scripts, buyer emails, season reports from aggregated data, plain summaries of EU requirements and Excel for collection records.

Stays with people: farmer identity, plot data, quality grading and contract terms.

The Ugandan detail: Uganda's coffee exports were reported at about 8.4 million bags worth US$2.4 billion in the 12 months to October 2025. The EU Deforestation Regulation applies from 30 December 2026 for large and medium companies and 30 June 2027 for most micro and small ones. Farms must be mapped, so a great deal of farmer data now exists.

The Independent reported in February 2026 that registration was funded for 900,000 farms in one phase and 780,000 in another, with the data held in a National Data Warehouse hosted by NITA-U. It also quoted a farmer who gave his details and never received proof of registration. Training should teach field teams to explain what is collected and why, and to keep names and coordinates apart from any AI step.

Manufacturing

Good first uses: turning approved procedures into shift briefings, handover templates, downtime analysis in Excel, supplier letters and job descriptions.

Stays with people: technical sign-off, safety decisions, quality release and commercial terms.

The Ugandan detail: at a February 2025 dialogue, the trade minister put manufacturing at 16.5% of GDP and over 3 million jobs, and the Uganda Manufacturers Association chairman put capacity use at 53%. Outages are a real planning issue, so a restart-after-outage briefing is a useful first exercise. Shift patterns also mean training has to run in short blocks.

NGOs and the refugee response

Good first uses: donor narratives from aggregate figures, proposal outlines, minutes of coordination meetings and plain versions of technical guidance.

Stays with people: every impact claim, every figure and every decision about people.

The Ugandan detail: UNHCR counted 1,999,576 refugees and asylum-seekers in Uganda on 31 March 2026, the largest refugee population in Africa, with 120 partners in the response. Only 14% of UNHCR's 2026 requirement was funded at that date, and it had closed offices in Arua, Moyo and Kyaka II. Smaller teams are carrying the same reporting load, which is why AI is tempting and why a wrong paste is so dangerous. Our guide to AI training for NGOs goes deeper on proposals and monitoring.

Government-linked bodies

Good first uses: memos, minutes, public notices, summaries of published reports and budget formulas.

Stays with people: classified files, personal records and final decisions.

The Ugandan detail: demand is visible. More than 1,000 secretaries and administrative professionals attended a two-day AI workshop at the Civil Service College in Jinja in May 2026. The new National Data Strategy calls for anonymisation and pseudonymisation under the 2019 Act, so masking is a skill public officers should practise now.

English, Luganda and the languages staff really use

Keep AI work in English, then have fluent people own every other language. Article 6 of the Constitution makes English the official language and Swahili a second official language for uses Parliament prescribes.

The Constitution gives Luganda no official status. A 2015 article in The Conversation described it as the most widely spoken indigenous language and the most widely spoken second language alongside English. That article is over a decade old, so treat it as a guide.

Here is what vendor lists showed on 8 November 2026.

Chat assistants are a separate question. Google's Gemini page lists Swahili and does not list Luganda. Microsoft's Copilot language page does not list Luganda, and lists Swahili only for voice on personal accounts. Absence from a list means the vendor makes no promise. Microsoft says Copilot may show an error in an unsupported language. Other tools may still answer, so test how well.

How good is it, in numbers?

Sunbird AI, a Ugandan organisation, published the best comparison we found as a preprint in October 2025. It tested translation across 31 Ugandan languages using a character-match score (chrF), where higher means closer to a human reference. A high score does not mean a text is ready to publish. Sunbird built both the test and one of the models, and we saw no peer review, so read it with that in mind.

Across all 31 languages, the averages fell sharply: Sunflower-32B 0.357, Gemini 2.5 Pro 0.301 and GPT-4o 0.235. Two conclusions follow. Luganda does far better than the average Ugandan language in mainstream tools, and a team in Lira or Soroti cannot assume what works in Kampala will work for them. These model versions have since been replaced, so run your own check.

Sunbird's Sunflower models are worth including in any Ugandan course. The Ministry of ICT marked their launch in October 2025, and Sunbird's documentation lists chat and translation for the languages in the table, with speech support for some. Apply the same data rules to it as to any outside tool.

A twenty-minute Luganda drill

Take a real, non-sensitive notice from your work, such as a branch closing early. Ask two or three tools to translate it into Luganda. Fluent staff score each version for accuracy, tone and respect, including whether it would embarrass the organisation if read aloud in a market or on radio. Repeat for your other main working languages.

Mobile money and connectivity: what to plan around

Most staff will reach AI through a phone on a paid bundle, and the network or the power can stop without notice. Training should assume this.

Shutdowns are a planning item for any multi-week programme. The UCC ordered internet access off from 18:00 on 13 January 2026. Connectivity was disrupted until 18 January, and OONI measured WhatsApp blocking on MTN, Airtel and Roke Telkom after restoration. Avoid building a course that depends on one WhatsApp group. Use email as the backup and give people a PDF pack they can keep.

Starlink is newly licensed. The UCC signed its operating licence on 15 May 2026, and reports describe enterprise resale from August and consumer ordering from September 2026. The reported licence conditions include registering devices with the UCC, so check any kit before relying on it.

Paying for bundles. If the employer funds data, buying a bundle directly is cleaner than sending cash, since the 0.5% withdrawal excise and fees reduce what arrives. Text prompts cost little. File uploads, images and voice usually cost more, so teach staff to paste a short extract in place of a whole report.

Which free tools to use from Uganda

Gemini, Claude and the OpenAI API all list Uganda as supported, as of 8 November 2026. Microsoft says Copilot is available in over 170 markets and does not name them on that page, so test it on your own accounts.

We teach the free version of every feature and do not promote one tool.

Six prompts your team can try

Each one uses the four Cs we teach: Clarity, Context, Constraints and Customisation. Replace the square brackets with non-sensitive details.

1. Classify your columns before you paste anything

I am writing an internal AI rule for a Ugandan [type of employer]. Here are only the column names from a staff spreadsheet: [column names]. Section 9 of Uganda's Data Protection and Privacy Act lists "religious or philosophical beliefs, political opinion, sexual life, financial information, health status or medical records". Mark each column as special personal data, other personal data, or not personal data, with a one-line reason. Where you are unsure, write [ASK DPO]. Do not guess.

2. Reconcile mobile money collections in Excel

I have an Excel sheet of mobile money collections with columns: Date, Transaction ID, Amount (UGX), Fee (UGX), Branch Code, Invoice Ref. Give me formulas to match each row to an Invoices sheet, flag amounts that differ by more than UGX 1,000 and total by branch. Explain each formula in one sentence. I use basic Excel. Include an alternative if my version has no XLOOKUP.

The figures stay in your sheet, because the tool only needs the column names.

3. Explain farm mapping to a coffee farmer

I work for a coffee exporter in Uganda. Write a two-minute script in very simple English that field officers can read to smallholder farmers before recording a farm's GPS location. Explain what is collected, why EU buyers ask for it, who will see it and how to ask questions. Use only these approved facts: [approved facts]. Keep sentences under 12 words. Then list any words that may be hard to translate into Luganda or Runyankole.

A fluent colleague then writes the local language version.

4. Restart briefing after a power cut

Using only the approved procedure below, write a five-minute briefing for the next shift at a food factory on restarting [line] after a power cut. Give the checks in order, the person who signs off each one and two questions to ask the crew. Do not add any technical step that is not in the procedure. Mark gaps with [CHECK]. A supervisor will approve this. Procedure: [non-confidential extract]

5. Customer message for a small screen

Rewrite this approved message about mobile money charges in plain English in under 160 characters. Then give a version an agent can read aloud in under 40 words. Keep every figure exactly as written, invent nothing and avoid idioms so a colleague can translate it into Luganda. Message: [approved text with no customer data]

6. Check the draft against the source

Compare the draft with the source below. Make a table with three columns: the claim or number in the draft, where it appears in the source, and "not in source" if it does not appear. Do not rewrite anything. Draft: [draft]. Source: [source].

Start with prompts 1 and 6, because they guard against a wrong paste and a wrong claim.

What a good programme looks like

A good programme starts with compliance, runs over weeks and uses your own work. This is the shape we would recommend.

1. A written scoping brief. Which tasks, how many people, which sites, which devices and what language needs.
2. The PDPO check. The four questions from earlier, answered before the first session.
3. A two-page AI policy. Approved tools, the paste test, human sign-off and who to ask. Your DPO should be in the room.
4. Four to six live sessions of 60 to 90 minutes. Foundations and the four Cs, safe data handling, your own tasks, then Word, Excel and Copilot.
5. The language drill, run by fluent staff, ending in written rules for review.
6. Champions in each branch or district. Two named people who keep going after the trainer leaves.
7. Honest measurement. Time one task before and after, including checking time, and log the errors AI made.

Nobody becomes an expert in a few days, and a course should say so. Weeks of practice keep skills alive, as we explain in why corporate AI training does not stick. If you are going to tender, our note on what to put in an AI training RFP gives you the clauses.

Mistakes to plan for

• A course with nothing about NINs, mobile money or the PDPO.
• HR booking training before anyone has checked the register or the undertaking.
• Luganda output sent out because "it looked fine" to someone who does not read Luganda.
• All course contact on one WhatsApp group.
• A farmer list or payroll file pasted in "just to tidy the columns".

Where training will not help

Training cannot fix an organisation that has not decided who may use which tools. It will not make AI reliable in languages it handles poorly. It will not replace a data protection review for a system that makes decisions about people, and a lawyer should look at any such project.

Questions people ask

Does Uganda have a law on AI?

We found no AI-specific law as of 8 November 2026. A national AI and emerging technologies strategy was in draft, and the government launched a National Data Strategy in October 2026 as groundwork. The Data Protection and Privacy Act already applies whenever AI touches personal data.

Do we have to register with the PDPO to use AI?

The duty to register comes from collecting or processing personal data, whether or not you use AI. The PDPO told chief executives in August 2026 that collecting personal data without registering is a criminal offence. Check your entry and purposes before rolling out AI.

Can staff paste customer data into ChatGPT or Gemini?

They should not, in a public version. Consent, the special category for financial and health data, and the section 19 rule on processing abroad all come into play. Masked data, or an approved work tool with a recorded basis for the transfer, is the safer path.

Can AI decide who gets a loan or a job in Uganda?

Be very careful. Section 27 lets people challenge decisions based solely on automated processing, and it has exceptions, including some connected with contracts. Keep a named person responsible, and take legal advice before automating any such decision.

How well does AI handle Luganda?

Better than most Ugandan languages. In Sunbird AI's self-published October 2025 test, Gemini 2.5 Pro and Sunflower scored close together on English to Luganda and GPT-4o trailed. A fluent speaker should still check anything that goes out.

Do you run AI training in Kampala?

Training for Ugandan teams is available live online, or in person by arrangement in Kampala and regional centres. We are based in Nairobi and have no office in Uganda.

Sources

All sources last checked 8 November 2026.

Law and regulator

• Data Protection and Privacy Act, 2019 (Parliament of Uganda copy)
• Constitution of Uganda (Constitute Project)
• DLA Piper, Data protection laws: Uganda (modified 13 February 2026)
• DLA Piper Privacy Matters, PDPO decision on offshore entities (6 August 2025)
• Daily Monitor, "Over 35,000 entities unregistered" (19 August 2026)
• Eagle Online, PDPO on unregistered firms (19 August 2026)
• Afriwise, Registration and compliance for NGOs (2022)

AI and data policy

• UNESCO, Building blocks for AI governance in Uganda (24 February 2026)
• PC Tech Magazine, AI and emerging tech roadmap by June (2 April 2026)
• Tech Review Africa, stakeholder consultations on the draft (2 May 2026)
• Daily Monitor, National Data Strategy launch (October 2026)
• iAfrica, National Data Strategy and the AI plan (23 October 2026)

Connectivity, power and payments

• PC Tech Magazine, UCC Q2 2026 figures
• Charmar News, UCC Q1 2026 figures
• Kampala Post, data use and mobile money in Q1 2026
• Les Afriques, data prices from Cable.co.uk (2 October 2026)
• Tech Africa News, UCC tax study (22 June 2026)
• The Independent via allAfrica, mobile money levy (15 April 2026)
• OONI, social media blocking after the 2026 election
• Dignited, Starlink licence (May 2026)
• Space in Africa, Starlink live in Uganda (1 September 2026)
• World Bank, access to electricity, Uganda
• Daily Monitor, national grid collapse (June 2026)
• Eagle Online, UEDCL network plan (20 August 2026)

Employers and sectors

• Bank for International Settlements, Governor Atingi-Ego on Uganda's digital future (2026)
• Further Africa, coffee exports (3 October 2026)
• The Independent (Uganda), farmers and data (2 February 2026)
• Hogan Lovells, EUDR simplification package (15 May 2026)
• SoftPower News, manufacturing figures (27 February 2025)
• UNHCR, Uganda fact sheet, January to March 2026
• Nile Post, secretaries' AI workshop at the Civil Service College (16 May 2026)

Languages

• Google Cloud, Translation language support
• Microsoft, Translator language support
• Microsoft, supported languages and markets for Copilot
• Google, where the Gemini app is available
• Sunbird AI, supported languages
• Akera et al., Sunflower paper (October 2025)
• Ministry of ICT and National Guidance, Sunflower launch
• The Conversation, English and local languages in Uganda (5 November 2015)

AI tools

• Google, Gemini plans
• Microsoft, Copilot Chat with and without a licence
• Google, Gemini Apps Privacy Hub
• Anthropic, supported countries
• Anthropic, updates to consumer terms
• Claude pricing
• OpenAI, API supported countries
• ChatGPT pricing

Where FuKazee fits in

FuKazee is a training company based in Nairobi, with no office in Uganda. Our founder began training learners in 2022, and we have now trained more than 3,000 people in tech, the future of work and everyday AI.

Training for Ugandan teams is available live online, or in person by arrangement. We teach in English and rely on your fluent staff for the Luganda and other language work described above.

We assume nobody in the room knows AI yet, and we avoid jargon. We show the free version of every feature and do not promote one tool. Microsoft 365 and Copilot are part of the course, along with the four Cs, safe data handling and checking every output. Each programme is built on your team's own tasks, devices and rules, with your data protection officer involved from the start.

To discuss training for your team, tell us your sector, how many people, where they sit and what you want AI to help with. Email lead@fukazee.com