Section 71 limits decisions made solely by automated processing. Build the human checkpoint in from the first design, not after the security review.
Agents are good at the work around a decision: reading the documents, extracting the facts, checking them against rules, drafting the recommendation. The risk for South African organisations is letting the agent take the decision itself, where the law expects a person to be involved.
What section 71 says
Section 71 of the Protection of Personal Information Act 4 of 2013 says, in short, that a person may not be subject to a decision that has legal consequences for them, or affects them to a substantial degree, where that decision is based solely on the automated processing of personal information intended to build a profile of them: for example their performance at work, creditworthiness, reliability, location, health, personal preferences or conduct.
There are exceptions. An automated decision may be allowed where it is taken in connection with concluding or performing a contract and the person’s request has been met or appropriate measures protect their legitimate interests, or where a law or code of conduct governs it and sets out such measures. The appropriate measures must give the person an opportunity to make representations about the decision, and give them sufficient information about the underlying logic of the automated processing to do so.
This note is a design guide, not legal advice. Your legal and compliance teams decide how section 71 applies to your processes.
Five design rules we follow
1. Classify the decision before you design the agent
For every workflow, we list the decisions in it and mark the ones that could have legal consequences or substantial effect for a person: a declined application, a flagged claim, a changed limit, a disciplinary step. Those decisions get a human owner in the design, before any code is written.
2. Agents recommend, people decide
Where section 71 could apply, the agent prepares the case and a recommendation. A person with real authority makes and records the decision. “Solely automated” is avoided by design, not by a reviewer clicking approve on everything.
3. Make the human checkpoint meaningful
A reviewer who sees only “approve” and “reject” is not exercising judgement. Show the evidence, the rules applied, the confidence and the reasons for the recommendation, and make disagreeing as easy as agreeing.
4. Log the logic, not just the result
To give a person sufficient information about the underlying logic, you need to have recorded it. Every recommendation is logged with its inputs, the sources and rules relied on, the output and the person who decided.
5. Build the route for representations
Design how a person can query or challenge a decision, who receives it, and how the case is reopened. If that route only exists in a policy document, it will not work under pressure.
For any decision your agent touches, ask: if this person asked why, could we show them the logic, the evidence and the name of the person who decided? If not, the design is not finished.
These rules are part of our standard build. See the controls we build to.
Sources
- Protection of Personal Information Act 4 of 2013, section 71 (automated decision making).
- Protection of Personal Information Act 4 of 2013, section 72 (transfers of personal information outside the Republic).