Certification of AI Systems Used in Energy Governance – Detailed Explanation With Case Laws

1. Meaning

Certification of AI Systems Used in Energy Governance means the legal and technical process of testing, assessing and approving artificial-intelligence systems that are used by energy regulators, governments, utilities and other public institutions to make or support energy-governance decisions.

AI may be used for:

electricity-demand forecasting;

tariff analysis;

renewable-energy planning;

grid congestion management;

licensing support;

electricity-market surveillance;

infrastructure-risk assessment;

energy-consumption monitoring;

regulatory compliance;

fraud detection; and

environmental and climate-risk assessment.

The important point is that these systems may influence public power. Therefore, certification cannot be limited to ordinary software quality. It must also examine legality, fairness, transparency, accountability, cybersecurity and protection of rights.

2. Why Certification Is Necessary

An AI system used by an energy regulator may process enormous amounts of information and generate recommendations much faster than human officials.

However, AI can produce:

inaccurate predictions;

biased outcomes;

unexplained decisions;

incorrect risk classifications;

manipulated outputs;

cybersecurity vulnerabilities; or

excessive reliance on historical data.

For example, an AI system used to identify electricity consumers at risk of non-payment could incorrectly classify certain communities because its training data contains historical biases.

Similarly, an AI model used to recommend electricity infrastructure investment could favour areas that have historically received more investment.

Certification therefore creates a legal mechanism for asking:

Is this AI system sufficiently reliable, lawful, secure and accountable to be used for energy governance?

3. Risk-Based Certification

A useful approach is to classify AI according to its potential impact.

Low-Risk Systems

For example, an AI system that prepares internal statistical reports.

Requirements could include basic accuracy testing and documentation.

Medium-Risk Systems

For example, AI that recommends tariff structures or infrastructure priorities.

These systems require stronger validation, human review and auditability.

High-Risk Systems

For example, AI that substantially influences:

electricity licensing;

market enforcement;

network-access decisions;

major infrastructure approvals;

consumer restrictions; or

emergency energy measures.

Such systems should be subject to rigorous independent testing and continuing supervision.

4. Certification Criteria

A comprehensive certification framework should examine several areas.

Technical Accuracy

The system must perform reliably under normal and unusual conditions.

Data Quality

Training and operational data should be accurate, relevant and sufficiently representative.

Explainability

Regulators should be able to understand the important factors influencing an AI-generated recommendation.

Cybersecurity

The system should be protected against hacking, data poisoning and unauthorised modification.

Human Oversight

Officials must retain the ability to review, reject or override important AI recommendations.

Auditability

The system should preserve records showing:

data used → model version → output → human decision → final action.

5. Public Power and Administrative Law

The most important legal issue is that energy governance involves public power.

If a regulator uses AI to make or substantially influence an administrative decision, ordinary administrative-law principles do not disappear.

In Affordable Medicines Trust v Minister of Health, the Constitutional Court emphasised that regulatory powers must be exercised within lawful authority and that discretionary powers must be properly structured.

Similarly, Democratic Alliance v President of South Africa established the importance of rationality in the exercise of public power.

Applied to AI governance, a regulator should therefore be able to demonstrate:

legal authority → relevant information → rational methodology → lawful procedure → reasoned decision.

An official should not simply say:

“The algorithm decided this.”

The human decision-maker remains responsible for ensuring that the final exercise of public power is lawful.

6. Procedural Fairness

Section 33 of the South African Constitution protects the right to lawful, reasonable and procedurally fair administrative action.

The Promotion of Administrative Justice Act 3 of 2000 gives effect to this constitutional protection.

This creates an important question:

If AI substantially influences a regulatory decision, how can the affected person challenge or understand that decision?

Certification should therefore consider whether the system can produce sufficient information for:

reasons;

review;

appeal;

correction of errors; and

independent investigation.

An AI system that cannot provide a meaningful audit trail may be unsuitable for high-impact governmental decisions.

7. Energy Regulation

The Electricity Regulation Act 4 of 2006 provides the foundation for electricity regulation in South Africa, with NERSA performing important regulatory functions.

AI may increasingly support regulatory tasks such as:

analysing licence applications;

monitoring market behaviour;

forecasting demand;

examining tariff information;

detecting irregularities; and

evaluating system reliability.

Certification should ensure that AI supports, rather than unlawfully replaces, statutory decision-making authority.

Where Parliament has assigned a legal power to a particular authority, that authority cannot simply transfer the ultimate responsibility to an algorithm without proper legal authority.

8. Accountability and Responsibility

A major principle should be:

AI assistance does not eliminate human responsibility.

The legal framework should identify responsibility among:

AI developers;

energy utilities;

regulators;

government departments;

system operators; and

individual decision-makers.

Contracts and procurement rules should also require vendors to provide:

technical documentation;

security information;

model-change notifications;

audit access;

incident reports; and

appropriate indemnity arrangements.

9. Electricity Reliability

AI used for energy governance can influence decisions affecting critical infrastructure.

Eskom Holdings SOC Ltd v Vaal River Development Association is relevant by analogy because the Constitutional Court considered electricity supply and the wider consequences of electricity interruptions.

This suggests that AI governance should include reliability and resilience testing.

Before an AI system is used for important energy decisions, authorities should examine:

what happens when the AI fails;

whether incorrect data produces dangerous results;

whether human operators can intervene;

whether there is a backup system; and

whether decisions can be reversed.

10. Environmental and Climate Governance

AI may also influence environmental decisions involving energy infrastructure.

In Fuel Retailers Association of Southern Africa v Director-General: Environmental Management, Mpumalanga, the Constitutional Court emphasised integrated consideration of environmental and socio-economic interests.

In Earthlife Africa Johannesburg v Minister of Environmental Affairs, climate-change impacts were held relevant to environmental decision-making concerning major electricity infrastructure.

These cases are analogical rather than direct AI cases. They demonstrate that technological tools cannot remove the underlying environmental responsibilities of decision-makers.

If AI is used for climate-risk analysis, certification should therefore test whether its assumptions properly account for relevant environmental information.

11. Bias and Energy Justice

AI certification should also examine discriminatory effects.

Energy governance affects different communities differently. An AI model trained on historical electricity-consumption or payment data may reproduce existing inequalities.

Certification should therefore involve:

bias testing;

demographic impact assessment where legally appropriate;

representative datasets;

independent validation; and

mechanisms for correcting discriminatory outcomes.

This is especially important where AI affects electricity access, service prioritisation, enforcement or infrastructure allocation.

12. Cybersecurity and Data Protection

Energy-governance AI systems may process sensitive operational and consumer information.

Security requirements should address:

unauthorised access;

manipulation of training data;

adversarial attacks;

ransomware;

model theft;

false sensor information; and

malicious alteration of outputs.

South Africa's Cybercrimes Act 19 of 2020 forms part of the relevant cybersecurity framework, while data-processing obligations may also arise under the Protection of Personal Information Act 4 of 2013.

13. Continuous Certification

AI systems can change after deployment through:

software updates;

retraining;

new datasets;

changed algorithms; or

integration with new systems.

Certification should therefore not always be permanent.

A suitable model is:

initial certification → deployment → continuous monitoring → periodic audit → material-change assessment → recertification.

Serious incidents should trigger an immediate review.

14. Conclusion

Certification of AI Systems Used in Energy Governance is necessary because AI can increasingly influence decisions involving electricity markets, infrastructure, consumers, climate policy and regulatory enforcement.

An effective certification framework should include:

risk classification → technical testing → data-quality assessment → bias testing → cybersecurity → explainability → human oversight → administrative-law compliance → independent auditing → continuous monitoring.

The cases Affordable Medicines Trust, Democratic Alliance v President, Eskom v Vaal River Development Association, Fuel Retailers Association, and Earthlife Africa provide useful legal principles, although they are not direct cases concerning AI certification.

The central principle is that automation cannot replace legal accountability. If an AI system assists an energy regulator, the responsible public authority must still be able to demonstrate that its decision was lawful, rational, procedurally fair and based on relevant evidence. Certification therefore acts as a bridge between AI governance and traditional energy regulation, ensuring that technological innovation remains consistent with constitutional administration, electricity reliability, environmental protection and public accountability.