The growing risk of betting everything on public LLMs
Many South African companies are walking blindfolded into an unhealthy (and unsustainable) dependence on public large language models (LLMs) without understanding the commercial and governance risks, according to software development specialist, Global Kinetic’s CEO, Martin Dippenaar.
Global Kinetic builds software for highly regulated clients and Dippenaar says while financial institutions and other large enterprises are (rightly) anxious about privacy, many are still underestimating how quickly public AI services can become a single point of failure for their businesses.
“This is not just another SaaS decision. If you are 100 percent dependent on a single public LLM provider and you don’t have a backup plan, you’re running an uncontrolled experiment on your own business,” he warns.
An insidious token tab
Dippenaar’s argument begins with the question of cost. He describes his own use of Anthropic’s Claude as a “money furnace”, referencing US academic author and podcaster, Scott Galloway, who originally used the term to describe xAI.
Dippenaar says the cost of his personal usage has climbed exponentially and that when he asked the answer engine to predict the company’s future usage, based on the actual usage of our most proficient AI developer, the machine calculated the AI bill would rapidly climb towards US$10,000 a month as usage spreads across teams.
He also highlights more subtle forms of cost escalation.
“When users start a new chat session, the interface may default to a higher‑end, more expensive model. If they don’t notice the switch immediately, they can burn through their token allocation within an hour. They also can’t downgrade the model mid‑session without abandoning the conversation and starting again,” he says.
Dippenaar’s concern is not that these services are overpriced in absolute terms, but that most CIOs and CFOs have not yet modelled what happens when hundreds or even thousands of staff weave the tools into their daily workflows.
Lessons from .Net
Looking beyond cost, Dippenaar believes the major AI providers are deliberately building a new kind of platform lock‑in, echoing earlier eras of enterprise software.
He points back to the days when Microsoft entrenched its position in the enterprise by owning the developer ecosystem around .NET. “If they owned the developers, they owned the stack,” he says.
Organisations are encouraged to feed their entire knowledge base into the vendor’s ecosystem. From rate cards and proposals to engagement models, governance documents, financial data and historic project files – all fed into retrieval‑augmented generation (RAG) pipelines.
Once all of that is in place and working, the system can answer questions in the language and context of the business itself. The problem, Dippenaar says, is what happens next.
“Once that’s working seamlessly, it’s brilliant. But now imagine trying to move. You’re not just swapping one model for another. You’re rebuilding the knowledge substrate of your business in a different environment. That’s not a switch. That’s a re‑platforming,” he explains. “Boards shouldn’t be allowing such deep concentration on a single foreign‑controlled platform without a defined exit path.”
He also says far too few South African enterprises have interrogated this risk and says he has yet to see serious scenarios‑planning exercises asking the difficult questions facing AI’s future – both locally and abroad.
Rogue agents, escaped sandboxes, and geopolitical threats
The media is awash with existential threats facing the future of AI and how companies will access and use it – and for good reason.
Dippenaar, however, remains sceptical of official narratives around recent incidents in which powerful models reportedly collaborated to escape their sandboxes and probed external systems such as Hugging Face. He also questions why some of these stories have been made public now, just as AI firms jockey for higher valuations and potential IPOs.
In addition, the geopolitical threat facing the technology is growing as tensions between Washington and Beijing mount. A BCG report has warned that: “Diverging strategies between the AI superpowers—the US and China—are creating two increasingly incompatible tech stacks. For companies, the window for mixing the two may end sooner than expected.”
Dippenaar’s message to CIO and boards is not that AI should be abandoned, but that these systems introduce a new kind of security and control risk that is still poorly understood. “We are giving extremely capable systems tools, credentials and network access. You may not yet fully understand the risk surface, but you will be exposed to it,” he says.
Towards a more private AI future
Looking ahead, Dippenaar says the AI industry will likely move in cycles between centralisation and decentralisation, as it has in previous computing eras.
He believes the current rush to public LLMs will be followed by a swing towards more private and controlled AI environments, particularly in sectors that cannot afford uncontrolled data leakage or platform dependence.
He envisages a hybrid model becoming the safe bet. A controlled internal AI layer handling sensitive work, governance and core IP, with public frontier models used selectively for specific, high‑value tasks under explicit budget and approval controls.
“The economics are moving too fast to say everything will go private,” Dippenaar says. “But if you don’t start designing your exit ramps and your Plan B now, you’ll discover too late that your entire business is sitting on someone else’s platform, and on someone else’s terms.”
