Governing AI Under King V
South African boards working through their first King V reporting cycle are discovering that the technology principle is harder to discharge than it is to summarise. The recommended practices read as though they describe a destination. In practice they describe a record — what the governing body asked, what it was told, what it decided, and why that was proportionate to the organisation it governs. This piece sets out how to build that record, and where the law, rather than the Report, sets the real floor.
King IV's Principle 12 already placed the governance of technology and information with the governing body, supported by recommended practices on setting direction, delegating, overseeing and obtaining assurance. King V Principle 10 consolidates that accountability and makes two things materially more explicit: the treatment of data as a governed asset across its lifecycle, and the treatment of artificial intelligence and automated decision-making as a board-level accountability rather than a procurement question.
For most boards the correct starting point is therefore a gap analysis, not a new framework. A board with an evidenced Principle 12 trail should be identifying the delta — principally data classification and lifecycle, AI accountability and human override, and third-party and cross-border assurance — and extending what it already does. Boards told they face an entirely new obligation commission new committee architecture. Boards told they face a strengthened one close the gaps, which is cheaper and usually more candid.
Assurance: considered, calibrated, explained
The recommended practices contemplate that the governing body should consider periodic assurance on the effectiveness, compliance and ethics of the organisation's data and information handling and of its technology activity. "Should consider" is doing real work in that formulation, and boards are poorly served by advisers who read it as a universal requirement to commission an external AI or cyber security audit.
What assurance is appropriate depends on the organisation's circumstances and its significant risks, read with the principle as a whole. For a company whose only material automated tool is a staff rostering algorithm, a combined assurance view drawing on management reporting and internal audit may be entirely adequate. For a lender running credit scoring models, or an insurer running automated claims triage, it is unlikely to be. The board's record should show which significant risks it identified, what assurance it obtained over each, and why the two match.
One oversight item is worth isolating because it is routinely missed: minimum assurance requirements from outsourced technology providers over controls on significant risks. That is a contracting obligation as much as a governance one. If the operator agreement gives no audit right, no reporting obligation and no access to third-party attestations, the board cannot obtain the assurance the principle contemplates, and no amount of board-level diligence will manufacture it after signature.
Where the law sets the floor: POPIA section 71
Principle 10 is wider than section 71 of the Protection of Personal Information Act 4 of 2013. The principle reaches the acquisition, development, use and distribution of technology across the organisation. Section 71 is a specific right not to be subject to a decision based solely on the automated processing of personal information intended to create a profile, where that decision results in legal consequences for the data subject or affects the data subject to a substantial degree.
On the Information Regulator's only published comment on the provision — a short passage in its Guidance Note on Direct Marketing, issued in the direct-marketing context rather than as standalone guidance on section 71 — the right is engaged where the decision has legal consequences or substantial effect, the decision is based solely on automated processing, and that processing is intended to produce a profile of the person.
The Guidance Note is expressly advisory, and POPIA and the Regulations prevail over it in the event of inconsistency, so it cannot be presented as authority on the meaning of section 71. And the hardest practical question under "solely" is the reviewer who approves the model's output without independently assessing it: a human who rubber-stamps does not take the decision out of the subsection.
The exceptions are conditional, which is the part most often missed.
Section 71(2)(a) applies where the decision has been taken in connection with the conclusion or execution of a contract, and either the data subject's request in terms of the contract has been met, or appropriate measures have been taken to protect the data subject's legitimate interests. The distinction matters. Where the request has been met — the credit was granted, the policy issued, the application approved — the subsection is satisfied without more. Where it was not, the responsible party is thrown onto the appropriate measures limb, and section 71(3) then prescribes what those measures must do: give the data subject an opportunity to make representations about the decision, and require the responsible party to provide sufficient information about the underlying logic of the automated processing to enable those representations to be made. The practical consequence is that the safeguards bite precisely where the automated decision went against the person. A deployment that declines applicants automatically, and offers neither a representation mechanism nor an explanation of the logic, does not comply.
Section 71(2)(b) applies where the decision is governed by a law or code of conduct in which appropriate measures for protecting data subjects' legitimate interests are specified. Section 60(4)(a)(ii) requires every code of conduct to specify such measures insofar as automated decision-making under section 71 is concerned. But the route is available only to a responsible party falling within the scope of such a code or such a law. An organisation does not come within it because it has bought a commercially available tool, or because its sector has a code it is not subject to.
No South African court and no published Regulator determination has yet construed section 71. The absence of authority is not comfort: a board cannot predict how "solely" or "to a substantial degree" will be read, and should calibrate to the stricter rather than the looser reading.
The half of Principle 10 that gets ignored
Most board papers on technology governance are risk documents. The principle is not. The recommended practices address sustainable value creation and the capture of opportunity from emerging, innovative and disruptive technologies, and they treat benefit commensurate with significant technology investment as an oversight item in its own right. A board with an AI risk register and no view on which beneficial applications the organisation is evaluating has discharged half the principle.
Three questions belong on the agenda alongside the risk items:
Which applications are we actively evaluating, and what benefit do we expect from each? An evaluation pipeline is itself evidence of direction-setting.
How will the benefit be measured, and when do we return to it? Approving an investment and never testing its return is the same oversight failure as approving a policy and never testing it.
Have directors had the opportunity to understand the tools? A short briefing and a live demonstration of the systems the organisation actually uses improves the quality of oversight more than another policy document, and it makes for better adoption decisions. A director who has used the model is a better overseer of it.
Conclusion
Principle 10 asks the governing body to govern technology rather than to receive reports about it, and the disclosure framework asks it to say how. Done properly, that produces something a board can use: a demonstrable, proportionate record of direction, oversight and calibration. That record is not a defence under POPIA and should not be sold to a board as one. It is the material from which the organisation's compliance is built and the board's diligence is proved, and the difference between a board that has it and one that does not becomes apparent at exactly the moment it cannot be created.
For assistance with a King V readiness assessment, a section 71 automated decision-making inventory, an AI governance framework, or POPIA operator contracting, contact an expert at SchoemanLaw Inc in Cape Town or Paarl.
https://schoemanlaw.co.za/services/commercial-law/
https://schoemanlaw.co.za/services/contract-drafting/
https://schoemanlaw.co.za/services/technology-law/
Nicolene Schoeman-Louw | SchoemanLaw Inc
Specialist Technology, Commercial and Contract Law
Release information
Issued on behalf of
SchoemanLaw Inc
Media contact
SchoemanLaw Inc
SchoemanLaw Inc