South Africa Needs AI Accountability Without a Permission Slip
Good rules should make accountability unavoidable without making experimentation exceptional.
Written By: Gleb Tsipursky
South Africa is entering an awkward but useful moment in its artificial-intelligence debate. Government withdrew its draft national AI policy after fictitious sources were found in the document and said it would rework the framework to establish credible national standards on ethical AI use. Meanwhile, employers are not standing still. A human-capital summit opening in Sandton on August 12 will put leadership, AI and finance automation on the agenda.
That gap between policy and practice should not trigger a rush to rebuild an even larger central rulebook. It should encourage a more liberal approach: hold people and firms accountable for harms, require transparency where decisions materially affect others, and leave room for decentralized experimentation everywhere else.
The distinction matters because AI is not one activity. A system drafting an internal meeting summary is not equivalent to one screening job applicants, generating financial disclosures or making a decision that changes a citizen’s access to a service. Treating all of those uses as if they present the same risk would raise the cost of experimentation precisely when South African firms need more productive investment and new ways to create value.
Rational Standard recently argued that headline economic growth can conceal weak productive investment and job losses. That is the right context for the AI debate. South Africa cannot regulate its way to productivity, but neither can it automate its way there by replacing judgment with software and calling the result innovation.
A workable framework should start with responsibility rather than permission.
For low-risk uses, firms should be free to experiment. Let a sales team test an assistant for call summaries. Let an engineer compare drafting tools. Let a small business use AI to organize customer questions. The discipline should come from competition, professional standards, data protection and the ordinary consequences of poor performance—not from requiring advance approval for every new workflow.
For high-impact decisions, the burden should be different. A company using AI in hiring, lending, insurance, safety, financial reporting or other consequential areas should be able to identify who made the final decision, what evidence was checked, and how an affected person can challenge a serious error. That is not a licence to operate. It is accountability after power has been exercised.
This approach also avoids a common organizational mistake: assuming that a policy document can substitute for management. Employees need to know which tools are approved, what data cannot be entered, when human review is mandatory and who has authority to stop a faulty process. Managers need to make it safe to report failures. Otherwise, strict rules can push experimentation underground while senior leadership receives a falsely reassuring picture of compliance.
The government’s withdrawn policy illustrates why epistemic humility belongs in AI governance. A framework meant to govern a technology associated with unreliable outputs was itself damaged by unverifiable references. The lesson is not that government should abandon standards. It is that no institution—public or private—should be treated as infallible simply because it is writing the rules.
South Africa can set clear floors without dictating every ceiling. Existing legal duties around privacy, discrimination, consumer protection, professional responsibility and corporate governance can do much of the work when paired with specific guidance for genuinely high-risk uses. Sector regulators can respond to concrete harms in their domains. Courts and oversight bodies can clarify responsibility. Firms can compete on better safeguards as well as better products.
That decentralized model is not laissez-faire indifference. It is a recognition that knowledge about AI risk is dispersed across workplaces, professions and industries, and changes faster than a single authority can continuously codify. Good rules should make accountability unavoidable without making experimentation exceptional.
As South African executives gather to discuss AI and human capital, they should resist two seductive shortcuts: that technology can replace managerial judgment, and that regulation can replace it too. The durable path is narrower and more demanding. Give people freedom to test useful tools, draw hard lines around consequential decisions, and make a named human being answerable when those decisions go wrong.
That is how South Africa can protect liberty, encourage innovation and keep responsibility exactly where it belongs: with people.
Gleb Tsipursky, PhD, a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook