Wisevoter
South Africa

Experts Call for AI Fraud Scoring Reforms

Lesedi Modibane discusses the limitations of automated decision-making in South African fraud prevention systems.

Updated on Oct. 27, 2026 in Artificial Intelligence

Isometric editorial illustration showing a grid of solid geometric blocks, with one tilted outlier, representing the complexity of AI fraud scoring.
Experts at the 2026 IT Indaba in South Africa urged organizations to improve human oversight in AI-driven fraud detection systems to prevent wrongful claim rejections. AI Illustration. Upload story photo >

Live Poll

Do you trust automated AI systems to fairly determine your credit or insurance claims?

At the 2026 IT Indaba, experts highlighted the challenges of using artificial intelligence to detect fraud within the financial sector. While AI models successfully identify the vast majority of fraudulent cases, their tendency to incorrectly flag legitimate claims remains a significant concern for South African organisations.

Why it matters

Automated fraud scoring often struggles to distinguish between criminal activity and honest customer behaviour, leading to legitimate claims being wrongly rejected. As these systems become more prevalent, organisations must balance technical efficiency with the legal protections afforded to consumers.

AI models currently capture 90 percent of fraudulent activity but wrongly flag five percent of all legitimate claims. These systems operate on models that assessed 100 000 total claims to identify a one percent baseline fraud rate.

The players

Lesedi Modibane

An expert in technology and data governance who spoke regarding the limitations of automated fraud scoring at the 2026 IT Indaba.

The details

Existing legislation, specifically Section 71 of the Protection of Personal Information Act, restricts fully automated decision-making where it significantly impacts individuals. Consequently, organisations are being urged to maintain rigorous decision registers and clear human oversight to ensure accountability in their AI deployments.

Timeline

  1. Fraud cases rose by 26 percent throughout 2024.

  2. R1.4 billion in fraud was prevented by automated systems during 2024.

  3. Lesedi Modibane addressed the IT Indaba on 27 October 2026.

The Tech Race

This discourse follows the regulatory precedents set by Section 71 of the Protection of Personal Information Act, which mandates human oversight for decisions with substantial impacts. It marks a departure from purely algorithmic management as South Africa prepares to integrate the Draft National AI Policy 2026.

The inaccuracy of automated flagging can lead to frustrating delays or the wrongful denial of legitimate credit and insurance claims for everyday consumers. Users should remain aware that they possess rights regarding automated decisions under national privacy laws.

The takeaway

Organisations should implement robust human-in-the-loop oversight to correct the inherent errors in automated fraud detection models. Consumers experiencing unexplained service denials should query if an automated system was used and demand a manual review of their claim.

Further reading

For more information on national standards, visit the Artificial Intelligence section.

Source note: This article includes information reported by CIO South Africa.

Live Poll

Do you trust automated AI systems to fairly determine your credit or insurance claims?