Rick Vanover | Vice President | Product Strategy | Veeam | mail me |
South African organisations are moving quickly from Artificial Intelligence (AI) experimentation to deployment. The latest South African Generative AI Roadmap 2025 highlights this shift. It found that 67% of respondents reported current GenAI adoption. This figure is up from 45% in 2024. Therefore, organisations are moving from planning to active use.
While this momentum is real, it raises the stakes. AI is not a magic layer that fixes weak processes or poor data. Instead, it amplifies what already exists. If data is incomplete, poorly governed or difficult to recover, AI will scale those weaknesses just as efficiently as it scales productivity. As a result, AI trust becomes a central business requirement.
As a result, 2026 is becoming a defining year. Organisations are no longer asking whether enterprise AI works. Instead, they are asking whether it can be adopted safely. In sectors such as financial services, healthcare and retail, the answer depends on data integrity. Specifically, organisations must ensure that data is trusted, explainable, protected and recoverable under pressure. In practice, unified data resilience and AI trust now form a shared operational foundation.
As AI investment rises, so does the cost of failure
The business case for AI is no longer speculative. AI tools are moving into production environments. These environments directly affect customer decisions, patient outcomes, fraud detection, operational continuity and revenue generation. In South Africa, this shift coincides with rising enterprise risk. Consequently, organisations can no longer ignore these risks.
The Information Regulator’s 2024/25 annual report highlights this trend. It shows a significant increase in security-compromise oversight under POPIA. In addition, legal analysis from late 2025 reported 1,947 breaches since April 2025. This figure represents a sharp year-on-year increase. Although AI does not create these risks, it makes them harder to contain. This is especially true when the data layer is already weak. Therefore, AI trust depends directly on data resilience.
At the same time, cyber risk is becoming more expensive. A 2025 South African ransomware study reported that the average recovery cost exceeded R24 million. This figure excludes ransom payments. While this matters across industries, it becomes critical where AI systems shape decisions. These decisions must be defensible and recoverable. Therefore, if the infrastructure underpinning AI cannot support trust at scale, organisations face serious consequences. They risk failed projects, operational disruption, regulatory penalties, and reputational damage. In this environment, AI trust is no longer optional.
AI trust depends on data resilience
Think of AI as the voice of data. When organisations understand, govern and secure their data, AI trust becomes sustainable. In addition, they must ensure data is recoverable and provably intact during incidents.
Only then can AI trust withstand internal review and external scrutiny. At that point, AI evolves beyond a productivity experiment. However, many organisations are moving too quickly. They invest in AI while underestimating the fundamentals that ensure reliability.
Clean data, clear ownership and strict access control are essential. Furthermore, organisations must implement logging, recovery testing, and explainability. These are not optional features. Instead, they define the operating conditions for trustworthy AI and stable AI trust. This issue is particularly important in South Africa. Data accountability is already well defined.
POPIA guidance on Section 22 security-compromise notifications sets clear obligations. Organisations must safeguard personal information. They must also notify the regulator when compromises occur. In financial services, regulatory expectations are even higher. The Prudential Authority and FSCA Joint Standard on cybersecurity and resilience requires demonstrable controls. Therefore, AI systems must be explainable and recoverable. Otherwise, they will not meet regulatory expectations for AI trust.
What failed AI trust looks like
The consequences of weak AI trust follow consistent patterns across sectors. However, the impact differs by industry. In financial services, the primary issue is auditability and continuity. If AI influences lending, fraud detection or risk decisions, organisations must maintain a complete data record. They must also recover that record after a disruption. Without this capability, compliance teams cannot validate outcomes.
In healthcare, the risks are more immediate. AI systems increasingly support diagnostics, triage and patient prioritisation. If the underlying data becomes corrupted, incomplete or inaccessible, patient safety is at risk. Therefore, failures extend beyond operational inconvenience.
In retail, AI trust depends on continuous system performance. Inventory visibility, pricing logic and customer recommendations rely on consistent data. Fulfilment processes also depend on reliable systems. When these systems fail, the impact spreads quickly. Revenue declines, customer experience suffers, and brand trust erodes. Ultimately, AI trust collapses when data resilience fails.
Across all three sectors, the conclusion is clear. When data resilience fails, AI trust fails with it.
What successful AI trust looks like
Organisations that succeed take a different approach. They do not start with the most advanced AI models. Instead, they prioritise data visibility, governance and recoverability. This approach is especially important in multi-cloud and SaaS environments.
Our recent poll highlights this challenge. Nearly 60% of organisations reported reduced data visibility. This reduction results from the growth of multi-cloud and SaaS systems. Consequently, leaders cannot govern or secure what they cannot see clearly.
The path forward requires discipline. Leaders must understand what data they hold. They must also know where it resides and who can access it. In addition, they need visibility into how data is used. Most importantly, they must ensure rapid recovery when disruptions occur. Organisations should also identify redundant, obsolete and trivial (ROT) data. They must tighten permissions and reduce system sprawl.
Ultimately, the real test of AI lies in resilience. It is not about producing impressive pilot results. Instead, organisations must detect AI-related risks. They must protect AI-linked data assets. They must also recover cleanly from errors or cyber incidents. This capability transforms AI from experimentation into core infrastructure and strengthens AI trust.
The leadership question
AI readiness is no longer just a technical issue. It is a leadership responsibility. South Africa’s AI adoption continues to accelerate. At the same time, expectations around accountability and resilience are rising. Therefore, leaders must treat AI as a business capability. This capability must rest on trusted and recoverable data in order to sustain AI trust.
Leaders who adopt this approach will be better positioned. In contrast, those who treat AI as an add-on risk failure. Weak foundations cannot support scalable AI systems. As a result, long-term success depends on governance and discipline.
Ultimately, AI success depends on execution. It depends less on how ambitious a pilot appears. Instead, it depends on whether organisations build strong controls. They must also ensure recovery discipline and full data visibility. In critical industries, this distinction is decisive. It determines whether AI scales safely or becomes a costly new source of risk, and whether AI trust is achieved or lost.