South Africa: Directors' Duties for AI Governance Under King V

• South African boards are increasingly using AI for strategic decisions, and while the country lacks a standalone AI Act, the King V Code and existing company law now provide a framework for its governance.
• Directors can delegate tasks to assisted or augmented AI, provided they maintain supervision and ultimate responsibility, but delegation to autonomous AI is not permissible.
• AI systems have limitations, including potential bias from historical data, a 'black box' nature, and the absence of human qualities like empathy or intuition.
• A 2024 study raised concerns that AI in South African banking could perpetuate racial bias due to datasets reflecting historical inequalities.
• The evolving legal landscape, particularly with King V, creates an urgent legal and governance challenge, requiring directors to exercise independent judgment and act in the company's best interests, even with AI assistance.

Directors in South Africa cannot delegate their ultimate responsibility for company management, a principle that extends to decisions made with artificial intelligence assistance.

Artificial intelligence is increasingly shaping how boards of directors in South African companies approach decision-making. Directors are leveraging AI tools for critical functions such as evaluating financial information, forecasting market trends, and identifying potential risks before making strategic choices. However, this rapid integration highlights a significant gap: While South Africa does not yet have a standalone AI Act, the King V Code on Corporate Governance, effective January 1, 2026, and existing company law, provide a framework for governing the use of AI by boards, creating specific duties and responsibilities for directors.

This evolving legal landscape in South Africa is aligning with international developments. Countries like the European Union, South Korea, and Vietnam have enacted comprehensive AI legislation, while others, such as China, have introduced more targeted regulations. A recent academic paper by a company law researcher examined four key implications of AI use in South African boardrooms, concluding that existing local law offers minimal guidance on these complex issues. As AI becomes an integral part of daily business operations, this regulatory void is escalating into a substantial legal and governance challenge, underscoring the urgent need for clear legal rules to prevent a deepening legal vacuum.

AI can assist boards in various capacities, ranging from administrative support to advanced decision-making. The simplest form, assisted AI, handles administrative tasks like compiling meeting agendas, scheduling meetings, or drafting reports, allowing directors to focus on strategic matters while retaining full control over decisions. Augmented AI represents a more advanced application, capable of analyzing vast datasets, identifying patterns, and predicting potential outcomes for different business strategies. For instance, it could assess data to evaluate merger risks and propose optimal courses of action, with the board ultimately making the final decision.

The most sophisticated form, autonomous AI, can make decisions and execute tasks without continuous human intervention. Boards might, for example, authorize an AI system to make investment decisions within predefined parameters, independently monitoring markets and executing trades. Despite these advantages, research indicates significant limitations to AI's reliability. AI systems learn from existing data, meaning that inaccurate or biased input can lead to biased recommendations. This is particularly pertinent in South Africa, where historical inequalities are often reflected in available datasets. A 2024 study by the Centre of Excellence in Financial Services, for example, suggested that AI deployed in South African banking could inadvertently perpetuate racial bias.

Furthermore, many AI systems operate as a 'black box,' producing complex recommendations without clear explanations of how conclusions were reached, even for their developers. Another critical limitation is AI's inability to replicate essential human qualities vital for effective board leadership, such as empathy, intuition, and emotional intelligence.

South African company law permits directors to delegate certain functions to others, but it unequivocally states that they cannot relinquish their ultimate responsibility for managing the company. Directors remain fully accountable for all board decisions. This fundamental principle, according to the recent research, should equally apply to the use of AI within corporate governance. Directors may be permitted to delegate functions to assisted AI and potentially augmented AI, provided they maintain diligent supervision over the AI system and do not abdicate their core responsibilities.

However, delegation to autonomous AI is deemed impermissible. Allowing autonomous AI to make decisions would effectively grant the system decision-making powers, thereby enabling directors to relinquish their responsibilities, which is contrary to established company law. Directors are legally obligated to exercise independent judgment and to always act in the company's best interests. This means they cannot simply accept AI-generated recommendations without critical evaluation and independent thought, reinforcing that the ultimate burden of decision-making and its consequences rests squarely with the human board members.

Lawyers and compliance officers in South Africa should advise boards on the current legal vacuum regarding AI use, emphasizing that directors remain ultimately responsible for decisions made with AI assistance and must implement robust governance frameworks to mitigate liability risks. They should also monitor for impending regulatory developments in this area.

Source: Insights drawn from a recent company law analysis