AI Training in South Sudan: What Should Your Team Learn?

TL;DR:

• AI training South Sudan teams can use starts with everyday written work: donor reports, minutes, circulars, customer replies and spreadsheets. Free tools are enough to begin.
• Design for the network first. Sessions should be text-led, short and low on data, with material staff can download once and keep.
• Work in English, then have a fluent colleague write or check anything in Juba Arabic. Juba Arabic did not appear on the Microsoft or Google language lists we checked.
• As of 8 November 2026 we found no dedicated data protection law or AI law in force. A Data Protection Bill is being prepared. The Constitution protects privacy, and the Cybercrime and Computer Misuse Act, 2026 is law.
• NGOs and UN partners work under more detailed data rules than national law: humanitarian data guidance, the South Sudan Information Sharing Protocol and their grant clauses. Beneficiary data stays out of public AI tools.

Who this guide is for

This guide is for people who decide how staff in South Sudan learn to use AI. That includes HR and learning leads, country directors, compliance officers and heads of programmes.

It covers UN agencies, national and international NGOs, commercial banks, telecoms and mobile money teams, oil operators and their contractors, and government-linked bodies. It also covers South Sudanese professionals based in Nairobi, Kampala or elsewhere who work on South Sudan portfolios.

South Sudan already has people building with AI

Training should start from the talent already in the country. The University of Juba's AI Day, held on 20–22 August 2025, included generative AI workshops and student projects on tuberculosis detection from chest X-rays, traffic flow in Juba and an AI tutor for maths and science.

In April 2026, 17 young women took part in an AI app development bootcamp organised by the National Communication Authority through the Banat Initiative. One team built Maize-Sense, a maize disease tool designed to run on a phone without constant internet. Another wants to add voice notes in Juba Arabic and Dinka to its career guidance app. Go Girl ICT Initiative's YoGage AI research project is working on Juba Arabic itself.

The population is young, with a median age of 18.7 according to DataReportal. A training plan should give younger staff room to lead, and should treat local language knowledge as expertise the organisation needs.

The working conditions, as of November 2026

Connectivity is scarce and costly, and few people have electricity access, so course design has to change.

More capacity is planned. The government said in April 2026 that laying of about 2,700 km of fibre had begun and a national data centre was about halfway built. The US$53 million ASSIST project, launched in January 2025, covers grid extension in Juba, a mini-grid pilot in Yei and standalone solar in Upper Nile.

What teams can actually use AI for

Start with written and spreadsheet work that a person can check against a source. The table shows where that work sits in each type of employer.

Some of these sectors have live reasons to learn this now.

• Banks: thirteen commercial banks were on the national payment system in November 2025, and all commercial banks had to connect by 31 January 2026. New rules bring new staff briefings.
• Mobile money: the Electronic Money Regulation 2017 Amendment 2025 sets tiered limits, and front-line staff have to explain them. Radio Yei reported daily limits, set in South Sudanese pounds, of about US$250 for tier one accounts and about US$1,000 for tier two.
• NGOs: the NGO Forum lists 256 national and 107 international NGO members on its site, undated. The 2026 Humanitarian Needs and Response Plan asks for more than US$1 billion, so proposal and reporting work is constant.
• Oil: petroleum makes up about 79.5% of projected revenue in the draft 2026/27 budget tabled on 8 September 2026. Operators and contractors produce long procedures and handover records, which suit the first uses above.

How to design around connectivity and power

Good training here treats data and electricity as part of the budget. These are the choices we would make.

Text prompts are generally the cheapest way to use AI. File uploads, image generation and voice usually use more data, so teach staff to paste a short extract in place of uploading a whole report.

If a site has Starlink, confirm it is properly licensed. In July 2024 the regulator approved tariffs, required payment in South Sudanese pounds and warned that importing kits through unauthorised distributors is illegal.

English, Juba Arabic and the languages staff really use

English should be the working language for AI, and people should own every other language. Article 6 of the Transitional Constitution makes English the official working language and all indigenous languages national languages.

Here is what the vendors' own lists showed on 8 November 2026.

Bari and Zande were also absent from Google's list. A language appearing on a list says nothing about how well the tool handles your subject, so test it.

Juba Arabic is widely spoken as a lingua franca, and it is hard to build AI for. Go Girl ICT Initiative's executive director told a Juba workshop in June 2025 that communities pronounce the same words differently, which makes a standard training dataset difficult to build. She also warned that global moderation systems largely miss South Sudanese languages.

Three working rules follow from this.

1. Standard Arabic output is not Juba Arabic. Staff should treat it as a starting point at most, and never send it to a community unchecked.
2. Keep the AI step in English. Use it to make the message short and clear, then have a fluent colleague write the Juba Arabic, Dinka, Nuer or other version.
3. Do not use AI to read or sort community feedback in local languages. It may miss meaning, including a safety concern.

A useful workshop exercise takes ten minutes. Give the tool a short notice in plain English and ask for a Juba Arabic version. Fluent staff mark what is wrong. Your team then has its own evidence about how far to trust the tool.

What South Sudan's law says about data and AI

South Sudan had no dedicated data protection law in force in the latest sources we found, dated September 2026. Parts of the picture are still moving, so check before you rely on this table.

What "no dedicated law" means in practice

We found no data protection regulator to register with, and no statutory duty to run an impact assessment or report a breach to a data authority. A 2022 fact sheet recorded no regulator and no limits on sending data abroad. That sheet predates the 2026 Act. Reports on the new Act did not describe a transfer rule, and a January 2026 report on the new supervisory committee for the national data centre described no data localisation rule. We could not find the gazetted text of the Act to confirm this.

Privacy still has legal force. The Constitution and Section 96 apply whether or not a data protection statute exists. Someone whose information is mishandled may have a claim.

The Cybercrimes Act touches everyday AI use. Four points matter for a training room. The first three come from Radio Tamazuj's explainer.

• Knowingly publishing false information that causes panic, violence or economic loss can carry up to five years. An unchecked AI error in a public statement is a risk to take seriously.
• Sharing passwords, biometric data or multi-factor tokens without authority is an offence. Shared logins for AI accounts are a bad habit here.
• Creating fake social media or email accounts for deception is an offence.
• The Committee to Protect Journalists reports that Section 33 criminalises transmitting or receiving data without authorisation, with no public interest exemption. Staff should have clear permission before moving organisational data into any outside tool.

Authorities may be able to reach stored data. The sources describe this differently. Radio Tamazuj's explainer of Chapter IV describes court production orders, preservation of traffic data for up to 90 days and a judge's order for real-time collection. CPJ says Section 6(3) lets the National Communication Authority order internet service providers to give access to data it considers important for protecting users, without judicial oversight. Assume data held with a local provider can be requested, and ask counsel to read the Act for your situation.

A Bill is coming. Radio Tamazuj reported in December 2025 that the planned law would align South Sudan with Kenya's Data Protection Act and the African Union's Malabo Convention. Consent, minimal collection, security and access rights are sensible things to build now.

This section is general information. Take advice from a South Sudanese lawyer before setting policy.

Which rules apply to your team

Without one national statute, your obligations come from who you are and where your people sit.

For teams split between Juba and a regional office, apply the strictest rule to the shared work. One written policy is easier to teach than three. Our guide to an AI usage policy was written for Kenya, but its structure transfers well.

Donor and humanitarian data rules for NGOs

For most NGOs, the working rules on data come from the humanitarian system and the grant agreement. We did not find a published donor rule written specifically for AI use in South Sudan, so ask your grant manager in writing.

IASC guidance. The Inter-Agency Standing Committee's Operational Guidance on Data Responsibility (2023) covers personal and non-personal data. It defines sensitive data as data likely to cause harm to people, or to affect an organisation's ability to work, if disclosed. Sensitivity depends on context, so an aggregate table can still be sensitive.

The South Sudan Information Sharing Protocol. A March 2025 framework between the camp coordination and protection clusters names the November 2023 protocol as the basis for their data exchange. The IASC guidance says such a protocol should classify the sensitivity of specific data. Find the current version and teach from it.

ICRC handbook and AI policy. The third edition of the Handbook on Data Protection in Humanitarian Action has a chapter on AI and machine learning. An ICRC adviser's account of its 2024 AI policy says systems touching programme delivery or affected people need stricter management than back-office tools.

Check the current award. USAID stopped implementing foreign assistance on 1 July 2025, and remaining programmes moved to the State Department, so older templates may no longer describe your obligations.

A donor report must be true, so AI-drafted results need checking against the indicator tracking sheet, and any requirement to disclose AI use in a proposal should be followed exactly.

Our guides on AI training for NGOs and how to train NGO staff on AI go deeper on proposals and monitoring.

Which free tools work from South Sudan

Gemini, Claude and the OpenAI API all list South Sudan as a supported country, as of 8 November 2026. The OpenAI list we read is for its API, so test ChatGPT sign-up on your own accounts. Microsoft says Copilot is available in over 170 markets, excludes embargoed markets and does not name them on that page, so test that too.

Where your organisation already runs Microsoft 365, Word, Excel and Copilot Chat deserve time in the course.

Offices with very poor connections or very sensitive documents can look at open-source models running on a local computer. It needs a capable machine and steady power. Our guide explains how to run open-source AI models locally.

Seven sample prompts

Each prompt uses the four Cs we teach: Clarity, Context, Constraints and Customisation. Replace the square brackets with non-sensitive details. Adding "short plain text, no tables" saves data on a slow line.

1. Donor narrative from anonymised figures

Draft the narrative section of a quarterly report for a livelihoods project in [county]. I am a programme officer writing for an EU-funded grant. Use only these aggregate results: [indicator, target, achieved]. Do not add any number or claim that I have not given. Mark anything unclear with [CHECK]. Maximum 250 words, formal English, with a final sentence on the main challenge and the next step.

2. Preparing a notice for a human translator

Rewrite this community notice in very simple English so that a colleague can translate it into Juba Arabic for reading aloud on local radio. Use short sentences, one idea each, and no idioms. Keep the date, time and place exactly as written. Then list any words that may be hard to translate. Notice: [paste notice with no personal details]

3. Bank staff briefing from a public circular

Below is a public Bank of South Sudan circular. Write a one-page briefing for branch staff. State what changed, the date it starts, what staff must do differently and three customer questions with answers. Quote section numbers exactly. Where the circular is silent, write "not stated". Plain English. Circular: [paste public text]

4. Mobile money customer FAQ

I work in customer care at a mobile money provider in South Sudan. Using only the approved limits below, write six short FAQs that an agent can read out to a customer. Use simple English, under 40 words per answer, in a respectful tone. Do not invent fees or limits. Limits: [paste approved internal wording with no customer data]

5. Toolbox talk from an approved procedure

Using only the approved procedure below, draft a five-minute toolbox talk for a contractor crew at an oil facility. Include the main hazard, three controls and two questions to ask the crew. Do not add technical steps that are not in the procedure. The supervisor will check it before use. Procedure: [paste non-confidential extract]

6. Budget formulas in SSP and USD

I have an Excel sheet with columns: Budget Line, Approved (USD), Actual (SSP), Exchange Rate. Give me formulas to convert Actual to USD, calculate variance in USD and percent, and flag any line more than 10% over. Explain each formula in one sentence. I use basic Excel and I have not used pivot tables.

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

7. Checking the draft

List every factual claim, number, date, name and reference in the text below. For each one, say whether it appears in the source I have pasted. Write "not in source" where it does not. Do not correct anything yet. Text: [draft]. Source: [source].

Prompt 7 is the habit we would teach first, because it protects staff in every sector above.

What a good training programme looks like here

A good programme is spread out, low on data and tied to real tasks and a written policy. This is the shape we would recommend.

1. Scoping in writing. Which tasks, how many people, which sites, what devices and what connection. Email suits this better than a video call.
2. Policy first. A one or two page document covering approved tools, the traffic light list, human sign-off and who to ask. Without a national statute, this document carries much of the weight.
3. Four to six live sessions of 60 to 90 minutes. Foundations and the four Cs, safe data handling, the team's own tasks, then Excel and Word.
4. Practice on real, cleaned work. Staff bring a real task with the sensitive details removed.
5. A language session. Fluent staff test the tools in Juba Arabic and other languages, and write the rules for review.
6. Champions at each site. Two people who can answer questions when the trainer has gone.
7. Honest measurement. Time one task before and after, including checking time, and log the errors AI made.

Live online suits teams spread across Juba, field sites and regional offices. In person suits a group that shares a room and would lose too much to poor connections. A blend often works best: an in-person start, then short online follow-ups.

Nobody becomes an expert in a few days, and a course should say so. Practice over weeks is what keeps skills alive, as we explain in why corporate AI training does not stick. When comparing providers, our procurement checklist gives you questions to ask.

Training has limits. It cannot fix a site with no devices or no data budget, and it should wait if leaders have not agreed who may use which tools.

What goes wrong, and how to avoid it

Questions people ask

Does South Sudan have a data protection law?

Not as of the latest sources we found, dated September 2026. A Data Protection Bill is being prepared. The Constitution protects privacy, and the National Communication Act and the Cybercrime and Computer Misuse Act, 2026 contain related rules.

Is AI regulated in South Sudan?

We found no enacted AI law and no published national AI policy as of 8 November 2026. General law still applies to what people do with AI, including the Cybercrimes Act's offences on false information.

Can NGO staff use free AI tools for donor reports?

Yes, for drafting from aggregate, anonymised information, if your organisation permits it. Beneficiary-level data should never go in. Check your grant agreement and ask your donor contact where the rules are unclear.

Can AI translate into Juba Arabic?

It may give you something, but Juba Arabic was absent from the Microsoft and Google lists we checked. Treat any output as a rough draft for a fluent speaker to rewrite. Google Translate lists Dinka and Nuer, and these still need human review.

Can AI training work on a poor connection?

Yes, if it is designed for one. Text prompts use little data, so audio-first sessions, a PDF pack and phone-friendly exercises work on modest connections. Sites with very little access may need a shared room or in-person sessions.

Do staff need paid AI subscriptions?

Not to begin. Free plans are enough for learning, and organisations can decide on paid or work-managed accounts once they know which tasks are worth it. Work plans often carry stronger data terms, so check them before handling internal material.

Sources

All sources last checked 8 November 2026.

Law and policy

• Transitional Constitution of South Sudan (Constitute Project)
• National Communication Act, 2012 (ICT Policy Africa)
• Committee to Protect Journalists, "South Sudan's new cybercrime law ramps up threat of jail for journalists" (9 April 2026)
• Radio Tamazuj, explainer on Chapter IV of the Cybercrime and Computer Misuse Act 2026 (20 March 2026)
• Radio Tamazuj, explainer on offences and penalties (27 March 2026)
• ITWeb Africa, "South Sudan's Cyber Crime Act takes effect" (19 February 2026)
• Eye Radio, "Controversial cybercrime act takes effect amid free expression concerns" (18 July 2026)
• KICTANet, "South Sudan Ties Data Rules to Citizen Trust" (11 September 2026)
• Radio Tamazuj, "South Sudan to get data protection bill in 2026, says minister" (11 December 2025)
• Eye Radio, "Ateny outlines S. Sudan's digital transformation plans at Nairobi summit" (30 April 2026)
• Africa AI Policy Lab, Country Profile on AI Governance: South Sudan (version dated 7 October 2026)
• Eye Radio, South Sudan and AI investment, Kigali summit report (4 April 2025)
• DataProtection.Africa, South Sudan fact sheet (updated 24 May 2022)
• We Are Tech Africa, supervisory committee for gateways and the National Data Center (21 January 2026)
• GDPR Article 3
• DLA Piper, data protection in Uganda
• DLA Piper, data protection in Kenya

Connectivity, power and payments

• DataReportal, Digital 2026: South Sudan
• Internet Society Pulse, South Sudan
• Cable.co.uk, Worldwide Mobile Data Pricing (latest data 2023)
• Les Afriques, data price ranking using Cable.co.uk (2 October 2026)
• Developing Telecoms, Starlink launch and tariff approval (16 July 2024)
• World Bank, access to electricity, South Sudan
• Eye Radio, energy costs and the African Development Bank (19 February 2025)
• Eye Radio, ASSIST electricity project (17 January 2025)
• Radio Tamazuj, "Central Bank: Mobile money now legal tender" (July 2025)
• Radio Yei, "BoSS sets strict USD caps on mobile money" (1 September 2026)
• Eye Radio, banks and the national payment system (5 November 2025)
• The EastAfrican, South Sudan's time zone change (29 January 2021)

Employers, people and languages

• Radio Tamazuj, 2026/27 draft budget (8 September 2026)
• South Sudan NGO Forum, About
• Radio Tamazuj, 2026 humanitarian response plan (13 January 2026)
• University of Juba, AI Day programme (2025)
• Andariya, "Silicon Resilience: Inside Juba's First All-Female AI Revolution" (10 June 2026)
• Eye Radio, AI and local languages in South Sudan (20 June 2025)
• Google, new languages in Translate
• Google Cloud, Translation language support
• Microsoft, supported languages for Copilot

AI tools

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

Humanitarian and donor rules

• IASC Operational Guidance on Data Responsibility in Humanitarian Action (2023)
• CCCM and Protection Cluster cooperation framework, South Sudan (March 2025)
• ICRC, Handbook on Data Protection in Humanitarian Action
• ICRC Humanitarian Law and Policy blog, on the ICRC AI Policy (28 November 2024)
• FCDO Accountable Grant Low Value Arrangement Template (updated September 2025)
• Congressional Research Service, USAID overview (5 September 2025)

Where FuKazee fits in

FuKazee is a training company based in Nairobi, with no office in South Sudan. We have trained more than 3,000 people in tech, the future of work and everyday AI.

Training for South Sudanese teams is available live online, or in person by arrangement. We teach in English and rely on your fluent staff for the Juba Arabic 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, and so are the four Cs, safe data handling and checking every output. Each programme is built on your team's own tasks, connections and rules.

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