The inaugural UNITED Nations (UN) global dialogue on artificial intelligence (AI) governance, held in Geneva on 6 and 7 July, marked a genuine shift: For the first time, every country had a place in a universal UN forum on AI governance, debating trustworthy AI, accountability, capacity building and the global AI divide.
These are urgent priorities. Yet millions of people are already using AI in their daily work, raising a question that has received far less attention in Geneva: Who is teaching the workforce to use it responsibly and effectively?
Across public institutions, companies and professional services, employees use generative AI to draft, summarise, research and support decision-making – mostly without structured training, guidance or a proper grasp of its limitations.
The corporate world calls part of this ‘shadow AI’: AI use without the organisation’s knowledge, approval or safeguards.
It thrives where sanctioned tools and training are absent and creates real legal exposure: Confidential information entered into a public tool may be disclosed, and unverified output may not withstand scrutiny in a court, an audit or a disciplinary inquiry.
I see this gap every time I run an AI training session in Namibia. Most participants already use it at work. Almost none have been shown how to verify its output, what must never be entered into it, or when to set it aside entirely. The policy conversation assumes a level of competence that has not yet been developed.
GOVERNANCE NOT ALWAYS COMPETENCE
National AI governance and workforce AI literacy address different problems.
Governance determines how a country regulates AI, protects rights and establishes accountability.
Literacy concerns whether the individual using the technology can do so safely and critically.
A country can adopt a strong AI strategy, and an organisation an excellent AI policy. Both fail if employees cannot apply the rules.
A policy may prohibit disclosing confidential information, and training must explain what is confidential and how a tool could expose it.
A policy may require human oversight – training must show what meaningful oversight looks like and why merely reading an AI-generated answer is not sufficient.
Governance sets the rules. Literacy develops the competence to follow them.
THE GAP THAT REMAINS
The United Nations Educational, Scientific and Cultural Organisation has AI competency frameworks for students and teachers, the United Nations Development Programme trains government officials, the International Telecommunication Union runs an AI Skills Coalition, and the United Nations Institute for Training and Research offers a certificate in AI and professional skills.
But most of this work targets national capacity, education systems and broad access to courses.
The missing layer is practical, role-specific training within workplaces. A general course can cover prompting, bias and ethics.
It will not teach a lawyer to verify AI-generated legal authorities, a company secretary to safeguard board information, or a public official to document AI use in an administrative decision.
An AI-literate employee should be able to select the right tool, protect confidential and privileged information, recognise hallucinations and weak reasoning, verify important claims, maintain independent judgement, explain how AI contributed to the work, and know when not to use it.
The competence required varies with the consequences of error: A routine email is not recruitment, healthcare or legal advice.
INCLUDING THE WORKFORCE
The AI divide is not only between countries; it is also between people who can use AI critically and those who cannot.
The UN AI Governance for Humanity Lab in Valencia, whose early work examines how governance frameworks are implemented in practice, could support baseline competencies, train-the-trainer programmes, and methods for testing whether training changes workplace behaviour.
The UN cannot train every employee. Its role should mirror its human rights model: Set benchmarks and require member states to report transparently.
When the dialogue reconvenes in New York in 2027, that reporting should cover not only infrastructure and regulation but also the workforce actually trained.
Governments should embed AI literacy within workforce development plans. Professional bodies should define competent use of AI, as standards of care will not wait.
Employers should treat AI competence like any other professional competence: Define it, train for it, and make it safe to disclose.
You cannot govern what you do not understand. Governance will be implemented by people. If we are building a global system to govern AI, we must also prepare the people expected to work with it.
– Chisom Obiudo is an admitted legal practitioner of the High Court of Namibia, specialising in corporate and AI governance. She can be reached at chisomokafor11@gmail.com
