Practical AI training will help civil servants protect public information, verify AI outputs and use the technology responsibly.
By Chisom Obiudo
The Cabinet has directed the Office of the Prime Minister (OPM) and the Ministry of Information and Communication Technology (MICT) to develop an internal policy governing the use of artificial intelligence (AI) in the public service.
The Honourable Minister of Information and Communication Technology, Emma Theofelus, announced the directive at the ninth National ICT Summit in Ongwediva.
This is a welcome step toward providing civil servants clear guidance on how to benefit from AI while safeguarding government information and the people they serve.
The policy’s effectiveness will depend on whether staff understand its requirements and can apply them in their day-to-day work.
AI governance is the framework of rules, responsibilities and checks by which an institution oversees AI. It covers approving tools for specific tasks, protecting information, monitoring performance and responding to problems.
Implementing these arrangements requires trained staff, clear responsibilities and adequate resources.
AI literacy provides part of that practical foundation. It means having the knowledge, skills and judgement to recognise where AI can help, understand its limitations and use it responsibly.
An officer may know that confidential information must be protected yet not realise that uploading a document to an external chatbot sends its contents outside the government’s systems. Training should make that connection clear before the officer uses the tool.
International practice supports this approach. Article 4 of the European Union’s AI Act requires providers and organisations to deploy AI systems within its scope to take measures to promote AI literacy among staff and others who use those systems on their behalf.
These measures should reflect people’s roles, experience and the context of use. Namibia can draw on this principle when developing its own approach.
Why AI usage rules alone are not enough
Generative AI tools can draft text, summarise reports, translate material and create images. They respond to instructions, commonly called prompts, and may also process uploaded files.
These functions can appear in standalone chatbots or within software that users already use. A familiar application can therefore introduce new information-handling risks when you add an AI feature.
What happens to information entered into an AI tool depends on the service, the account settings and the terms governing its use.
Depending on those arrangements, the provider may store the information, permit authorised staff to review it or use it to improve its models.
Institutional agreements may restrict these activities and provide additional protections. Government should assess these arrangements and give officers clear guidance on which tools and accounts they may use, for which tasks and with what information.
The “guardrails” mentioned in the announcement should combine clear rules with technical safeguards to help prevent misuse.
Consider an officer preparing letters for PSEMAS applicants. To save time, the officer uploads a spreadsheet of applicants’ names and identity numbers to an unapproved chatbot.
Because the conversation is visible only through their account, the officer assumes the information remains private.
Yet uploading the spreadsheet sends its contents to an external service that the government has not approved to receive them. The concern is losing control over how that information is handled, even if it never becomes publicly available.
Practical training could show the officer how to draft a standard letter using an approved AI tool and fictional details. The officer would then enter each applicant’s personal information into an authorised government system. This demonstrates how AI can save time while keeping personal information out of the chatbot.
Training also helps officers recognise mistakes and know when to report them. A blanket ban, without guidance or an approved alternative, may drive AI use out of sight as staff turn to personal phones and accounts.
The policy should therefore make clear that its rules apply whenever AI is used for government work, regardless of the device or account. The duty to protect government information remains unchanged, even when the activity is harder to monitor.
What each level of the public service needs to know
AI literacy means having sufficient understanding of AI to use it responsibly in one’s work. It does not require every civil servant to learn to programme. Training should reflect each person’s responsibilities. An officer preparing a document needs to know how to use the tool and to check its output. A supervisor must assess the work, while an executive director must understand the risks of approving the system.
For officers, the starting point is to know which tools are approved, which tasks they may support, and what information may be entered. Officers also need to give clear instructions and assess whether the answer meets the task’s purpose.
A fluent response is not necessarily accurate or complete. An AI-generated summary may omit an exception that changes the meaning of a rule, while a draft letter may introduce a claim that the original documents do not support.
Training should therefore show officers how to verify AI outputs. Generative AI can produce fabricated facts and references, commonly called “hallucinations”. If a tool cites a section of an Act, the officer should consult the authoritative legal text to confirm that the section exists, is current, and supports the explanation.
Officers should check figures against official records and summaries against the original documents.
Asking the same chatbot to confirm its answer is not an independent check. Officers should also recognise when an answer contains unfair assumptions or requires specialist advice.
Supervisors need to consider both whether AI was appropriate for the task and whether the resulting work was properly reviewed. Drafting a standard acknowledgement letter carries different risks from assessing a person’s eligibility for a grant. Any proposed use of AI to support decisions affecting people’s rights or access to services requires closer scrutiny.
This should cover the legal basis, the process’s accuracy and fairness, and how an affected person can obtain an explanation or challenge the decision.
Effective human oversight requires a reviewer who can question the AI’s output and has the time, expertise and authority to correct or reject it. A signature alone offers little protection if the reviewer cannot explain the evidence or reasoning underpinning the final document.
Supervisors should know which checks were performed and whether further review is needed.
Executive directors and heads of agencies need to ask questions before approving a tool. What problem will it solve?
What information will it process, where will it be stored, and who will have access? Has the tool been tested for its intended use? Who will monitor its performance and respond to errors or disclosures?
Legal, information security and procurement specialists should inform the assessment. Institutional leaders remain responsible for ensuring that the necessary checks and safeguards are in place.
Five elements the government AI usage policy should contain
As OPM and MICT develop the policy, five elements would help civil servants understand its requirements and apply them in everyday work.
The first requirement is guidance on what is permitted. The policy should specify which AI tools and features it covers, including those built into software officers already use.
An accessible list should identify approved tools and accounts, the tasks they may support and the information they may process.
Approval should be specific: a tool authorised to help draft public notices may not be suitable for confidential records. Paying for a personal subscription does not make a tool approved for government work.
The second is practical training before officers use AI for official duties. Training should reflect the different responsibilities of officers, supervisors and institutional heads.
It should demonstrate permitted, useful tasks, such as improving a public notice or drafting a standard letter, and the checks each task requires.
Short exercises should assess whether participants can recognise sensitive information, spot an unreliable answer and decide when to seek advice.
Completing a course should demonstrate these skills. Refresher training should address changes to approved tools, policy requirements and risks identified through experience.
The third is clear rules for protecting information. The policy should prohibit officers from entering personal information, classified material, confidential Cabinet documents, or privileged legal advice into unapproved external AI services, whether free or paid.
Where AI processing of protected information is permitted, it should occur only in a system that has been assessed and expressly authorised for that purpose.
Officers should provide only the information necessary for the task. Training should also explain that removing a name does not necessarily make a document anonymous: an address, job title, or a combination of details may still identify the person.
The fourth is clear accountability mechanisms with effective human review. Officers should review the work they prepare using AI; supervisors should carry out the required review; and institutional leaders should provide the controls and resources needed for both.
The policy should specify when AI assistance must be recorded and what evidence of verification should be retained.
The greater the potential harm, the more rigorous the checks should be. Where AI supports a decision affecting a member of the public, the institution should be able to explain the basis of that decision and provide a means to challenge it.
The fifth is responsibility for implementation and regular review. Each office, ministry, and agency should designate a contact person to guide staff, coordinate training, and refer problems to the appropriate specialists.
The policy should specify who has authority to approve tools and their uses. A register of approved uses would help management track where AI is applied and who oversees it. Staff should know how to report accidental disclosures, serious errors or harmful outputs promptly.
Reporting procedures should encourage early correction while preserving appropriate accountability. The policy and the approved-tool list should be reviewed at set intervals and whenever significant changes or incidents warrant it.
Making the policy work in practice
AI literacy equips civil servants to use AI responsibly, but training must be supported by secure tools, clear approval procedures and access to advice. Officers need to know how to complete permitted tasks, check the results and recognise when a matter requires further review.
Institutions must provide the conditions that enable them to put that knowledge into practice.
Cabinet’s directive offers OPM and MICT the opportunity to develop the policy and its implementation arrangements jointly.
The policy should be accompanied by a funded training programme, clearly defined responsibilities and a timetable for implementation and review. Success should be assessed through everyday practice: can officers protect sensitive information, verify AI outputs and report mistakes promptly? These capabilities will turn written rules into reliable public service practice.
Chisom Obiudo is a legal practitioner specialising in corporate and AI governance. She is currently the chief legal officer at the Namibian Law Reform and Development Commission.
She previously served as a member of the National Artificial Intelligence Technical Advisory Committee, established by the National Commission on Research, Science and Technology (NCRST), with a focus on law and governance. She can be contacted at chisomokafor11@gmail.com.
