Three questions every South African board should put to its CISO this week
Who are the AI agents running under your corporate credentials, what did they do while nobody was watching, and who signed off on their access?
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Most South African board agendas this quarter include a slide on artificial intelligence. The slide is usually about adoption, pilots, use cases, maybe a new chatbot in the service desk. The agentic AI conversation is one layer below that, and it is the one no board slide yet covers. An autonomous AI agent is not a dashboard. It is a software identity that holds credentials, logs into systems, moves data, triggers transactions and sometimes creates records on the organisation’s behalf. Dozens of them are now running inside most mid-market South African enterprises, under the credentials of real employees, often without any identity governance record. The three questions below belong on every board risk agenda this week. Each is operational, specific, and the answers decide whether the organisation still has an audit trail at all.
CONTEXT AND BACKGROUND
The industry framing changed in October 2026 with the Hacker News analysis of the agentic AI velocity paradox: enterprises are deploying AI agents at machine speed on identity governance built for humans. Akamai’s research the previous month found nearly half of enterprise AI use already bypasses corporate security, creating what the firm calls massive shadow AI visibility gaps. IBM’s 2026 Cost of a Data Breach Report puts a figure on it: breaches involving an AI system cost on average six million US dollars.
INSIGHT AND ANALYSIS
The three questions are deliberately short. The first: who are the AI agents running under your corporate credentials? This is a non-human identity question. Most South African organisations can list every employee with privileged access by name. Very few can list every autonomous agent, which API keys it holds, which systems it can reach, and which human issued the credentials. The agent-side identity inventory is a governance item now, not a technology one.
The second: what did they do while nobody was watching? Identity governance platforms built for human workforces run on quarterly or monthly review cycles. An agent that exists for forty-seven seconds, elevates its privileges mid-run and completes a transaction will not appear in any such review. The audit trail of what the agent read, wrote and committed, under which authority, has to live inside the identity platform at agent speed. The National Institute of Standards and Technology launched its AI Agent Standards Initiative in early 2026, OWASP has published a Top Ten for Agentic Applications in which identity and privilege dominate, and CISA, with its Five Eyes partners, issued a joint advisory in May 2026 naming privilege as the foundational concern.
The third: who signed off on the agent’s access? If the answer is “the engineer who built the deployment”, the organisation has not delegated the review. A developer cannot be the independent reviewer of their own code, let alone the agent that code set loose. ITWeb’s analysis of POPIA-compliant AI agent deployments in South Africa argues that this separation is the specific control POPIA requires under Section 19.
IMPLICATIONS
The commercial consequence is already being priced in. Forrester’s 2026 Technology and Security Predictions report forecasts that twenty-five per cent of planned enterprise AI spend will be deferred from 2026 to 2027 as organisations rework their security architecture to catch up with the agents already running. South African boards signing off AI budgets this quarter without an identity governance line item are approving the exposure. For a board that cares about Section 76 of the Companies Act and about POPIA, the three questions above are the minimum due diligence.
CLOSING TAKEAWAY
The AI the South African board should ask about is not the chatbot on the website. It is the autonomous agent already running inside the company’s own systems, under the credentials of real employees, with no identity governance record of what it has done. Three questions, this week, put plainly to the CISO. Who are the agents, what did they do while nobody was watching, and who signed off on their access? The gap between the answer and the truth is the governance debt the organisation has accumulated while nobody was looking. South African boards that ask this week will fix it quietly. The ones that wait will fix it in front of the Information Regulator.
Author Bio
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments