South African insurers have reported using AI at a baseline level for general efficiencies. (Image source: iStock)
South African insurance providers are“quickly” catching up with banks in implementing artificial intelligence (AI) for their operations.
However, the country has yet to formulate a National AI Policy, which has seen insurers formulating their own AI frameworks in order to tap into the technology.
South Africa’s Draft National AI Policy was withdrawn earlier this year after it referenced completely fictitious, AI-generated citations, without proper verification.
The Department of Communications and Digital Technologies (DCDT) is still targeting the 2026/2027 financial year to finalise the AI policy.
As per the DCDT, an internal investigation − probing how the policy that featured as many as six AI hallucinated references was presented for public comment − is still underway, with two officials placed on precautionary suspension so far.
Communications minister Solly Malatsi told MPs in May that his department has begun to fix some of the fundamental errors that occurred. This included the move to seek experts in the AI field, governance and academia.
The advisory panel, he noted, will help the DCDT in the review and authentication of the references.
KPMG South Africa yesterday released its annual KPMG South African Insurance Survey,which details how local insurers operate in an environment characterised by geopolitical conflicts, AI advancements, evolving cyber threats and shifting customer expectations.
It hosted a virtual media briefing to unpack the analysis of the financial results of 29 non-life insurers, 15 life insurers and four reinsurers operating in South Africa.
Jan van der Merwe, associate director at KPMG South Africa, clarified that at a baseline level, AI is being used for general efficiencies, such as writing documents, e-mails and creating presentations, mirroring other industry verticals.
Some insurers are applying it in some of their core functions, he stated. For example, in non-life insurance, it is being applied to routine claims assessment, underwriting risks, fraud detection and client engagement.
However, AI isn’t being used for actuarial valuations or finance processes, he said. “I haven't seen an insurer applying an AI agent end-to-end to the valuation process.”
He noted that this is indicative of several things. “First, the implementation of AI isn’t necessarily that simple – it’s a complex process. The other point is that it’s expensive. The third, and more important point, is that AI is not trusted yet to just run by itself without human oversight. This shows the importance of having humans − what we call the human in the loop − providing oversight of what AI produces.”
Van der Merwe pointed to the other components in AI adoption and the speed at which it is moving: the cost of building solutions and the environmental footprint caused by computing power and energy use.
“My sense is that going forward, we might be in a position where insurers (companies at large) are required to publish the extent to which their AI footprint costs the company simplicity.
“Within the insurance industry, the AI winners will be those who adopt AI responsibly. It will be those who weigh up the cost versus the value, and lastly, the ones who bring in the human in the loop to make sure what the AI produces can be trusted.”
Hanyani Mabaso, associate director at KPMG South Africa, stated that South Africa’s AI policy vacuum means insurers must develop internal frameworks to manage risks related to data privacy and unauthorised data amendments. This, he believes, will guide them in their use of AI.
He referenced how a Google Gemini model − which was being trained in a safe, controlled environment − broke out of its sandbox testing environment during a cyber security evaluation.
“This example really demonstrates that it's important for companies to have some sort of framework and governance around the use of AI. Our insurers need to ask themselves who can amend the data that is either used or even produced by the AI.
“Insurers need to consider that AI has a lot of benefits, but there are a lot of responsibilities that come with that. So, the ask from an insurance perspective is quite big.”
Shifting to operational risk trends for South African insurers in comparison to global peers, Mabaso shared findings from a study comparing South African and UK insurers.
South African insurers, he said, are highly-concerned about cyber security, which they view as an interconnected risk. While AI offers new opportunities, cyber criminals can leverage the same technology without the constraints of laws or regulations to gain an advantage.
He pointed to recent reports of four local insurers suffering some form of data breach, whether it was a direct breach or through a third-party.
From a UK perspective, insurers’ concerns focused heavily on third-party risks, compared to South African insurers, he stated.
“When we looked at that stat, it did come a bit as a surprise at first because we know for a fact that South African risk committees are increasingly concerned about fourth-party risks, where they are worried about the third-party risks of their third-parties. It’s a risk that’s growing, both locally and globally.
“We anticipate that going forward, third-party risks for South African insurers are going to be elevated. This is mainly due to the interconnectedness of risks, from third-party risks, cyber risks and AI-driven risks.
“We think the insurers that will do well in the future are those that are going to invest in their risk management processes as an input in their strategic decision-making.”