An interview with Mtho Maphumulo from Adams & Adams Partners explored the growing use of artificial intelligence (AI) tools in drafting insurance claims and the unintended consequences this trend is creating for consumers, insurers, and dispute resolution processes.
The discussion began with the observation that AI tools are increasingly being used for everyday administrative tasks, including insurance claims submissions and responses. According to Maphumulo, this is no longer a rare occurrence in the insurance space. “We see it quite frequently lately,” he explained, pointing to a noticeable rise in AI-assisted documentation being submitted in claims processes.
One of the first challenges this trend has introduced is the significant increase in the volume of complaints being received by regulatory and dispute resolution bodies. Maphumulo referenced concerns raised by the National Financial Ombud (NFO), noting that there has been a shift in the nature of complaints compared to previous years. Where complaints were once relatively concise, often limited to “a page or two,” they are now frequently extended into “voluminous pages.” This shift, while reflecting more detailed submissions, has created an unintended burden on institutions that must review and assess these complaints. The practical consequence is that more time is required to understand each matter, which in turn can contribute to delays in the finalisation of disputes.
Beyond the issue of length, there is also concern about the quality and legal accuracy of AI-assisted submissions. In practice, insurers and legal practitioners are increasingly encountering arguments that appear structured but lack legal substance. Maphumulo explained that in some cases, policyholders use AI tools to challenge insurer decisions in ways that are not legally sustainable. He described some of these submissions as containing “lots of nonsensical reasoning that cannot really stand any sort of scrutiny.” While the intention of consumers may be to strengthen their claims or dispute outcomes, the reliance on AI without a proper understanding often results in arguments that are misaligned with established legal principles.
A further issue relates to the accuracy of legal principles generated by AI tools. Although the outright fabrication of case law appears to be less common, possibly due to increased public awareness of AI hallucinations, there are still frequent problems with outdated or misapplied legal concepts. Maphumulo noted that AI systems often produce “legal principles quoted, which do not exist or were abandoned many years back.” In other instances, legal rules are incorrectly framed or applied in contexts that do not align with South Africa’s current constitutional legal framework. This creates additional work for insurers and legal teams, who must carefully unpack and correct these inaccuracies before any meaningful resolution can take place.
The broader issue, according to Maphumulo, is not simply the use of AI, but the gap between what AI produces and what users understand. Many consumers rely heavily on AI outputs without properly reviewing or verifying the information provided. This is especially problematic in legal and insurance contexts, where accuracy and context are critical. He emphasised that AI can be useful, but only when the person using it has at least a basic understanding of the subject matter. Without that foundation, users are unable to properly assess whether the output is correct or relevant, which increases the likelihood of flawed submissions being used in formal disputes.
This gap in understanding also contributes to inefficiencies in the system. Instead of resolving disputes quickly, insurers and legal practitioners often find themselves spending additional time explaining fundamental concepts that should already be understood or verified. This, in turn, increases litigation and administrative costs. What could have been relatively straightforward claims processes became extended engagements requiring clarification, rebuttals, and repeated explanations of basic principles.
There are also legal risks for consumers when incorrect AI-generated information is submitted in formal proceedings. Maphumulo stressed that any correspondence exchanged between insurers and policyholders can become part of the official record if a matter escalates to bodies such as the National Financial Ombud or even the courts. “If you relied on incorrect outcomes from the AI tool, that will be part of the record,” he warned, noting that such submissions “may be quite detrimental to you as the policyholder.” Once included in the record, these documents can influence how a matter is interpreted and ultimately decided, potentially weakening a consumer’s position.
Despite growing awareness of AI’s limitations, behaviour has not necessarily changed in line with that awareness. Maphumulo observed that while people are generally aware that AI should not be relied upon blindly, many still proceed to use it without sufficient caution. There appears to be a tendency to prioritise convenience and speed over careful review and validation of the information produced. This disconnect between awareness and practice is one of the reasons why AI-related issues continue to appear in insurance disputes.
However, the discussion also highlighted that AI is not inherently negative when used correctly. There is clear value in its ability to assist with drafting, structuring information, and improving efficiency. The key requirement is responsible usage. Maphumulo suggested that consumers should, where possible, make use of insurance brokers. Brokers have a deeper understanding of insurance products and policy wording and are better positioned to provide accurate, context-specific guidance than AI tools alone. He explained that brokers are able to produce “more correct and more legitimate submissions” because they understand the underlying policies and industry practices.
For individuals who still choose to use AI tools, the advice is to do so responsibly. This includes conducting basic research before prompting AI systems, understanding the subject matter at a foundational level, and critically reviewing any output before it is used in a formal submission. AI should be treated as a supportive tool rather than a replacement for legal or technical understanding.
Ultimately, the interview underscores a growing reality in the insurance sector: AI is changing how claims are prepared and disputed, but it is also introducing new risks when used without proper oversight. As Maphumulo’s insights show, the challenge is not the technology itself, but the way it is applied by users who may not fully understand its limitations.
