The People Who Carry The Risk Are Not The Ones Who Can Reduce It

MIT surveyed 272 experts across 37 countries and produced, without meaning to, an accurate description of South Africa's position in the global AI system.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/the-people-who-carry-the-risk-are?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

Most research about artificial intelligence risk is written from the perspective of the countries building it. A new study is no exception, and that is precisely what makes one of its findings so uncomfortable to read from Johannesburg. The researchers set out to establish who is vulnerable and who is responsible. They were not thinking about South Africa. They described it anyway.

CONTEXT AND BACKGROUND

MIT FutureTech and the University of Queensland School of Psychology ran a three-round Delphi study in late 2025, engaging 272 international AI experts across 37 countries drawn from industry, academia, government and civil society, who assessed 24 risk categories over a five-year horizon. Under current trajectories without additional intervention, the experts judged 18 of the 24 risks more than ten percent likely to produce catastrophic outcomes, defined as more than one million deaths, more than a hundred billion dollars in financial losses, or comparable harms. Neil Thompson, who directs MIT FutureTech, observed that in other mature technology areas such as nuclear power or aviation, risks at that level would be treated as intolerable. The researchers caution that the per-risk probabilities are not independent and should not be added together, since the categories overlap and the scenarios may interact.

INSIGHT AND ANALYSIS

The finding that matters most here is structural rather than numerical. Experts assigned the highest responsibility for addressing AI risks to general-purpose AI developers and to governance actors including regulators, government leaders and standards bodies, while concluding that the users of AI models and the general public would bear the brunt of the harms yet have limited power to mitigate them. Andrew Gamino-Cheong, a participant in the study, described this as one of its biggest takeaways, being the consensus that the groups most vulnerable to the various AI risks were also the ones least responsible for them.

Peter Slattery, a research scientist on the project, drew out the mechanism, noting that those best positioned to reduce AI risks are not the ones most likely to suffer the consequences, that highly competitive development leaves labs and governments with limited incentive to slow down or invest sufficiently in safety, and that laws, treaties and other collective-action mechanisms may therefore become necessary. Read that description against South Africa’s actual position. We develop no frontier models. We sit on no standards body with meaningful weight over how those models are built. We are, in the study’s own terms, a user rather than a developer or a governance actor. That is not a grievance. It is a category.

I have previously written about this in an article on the rise of the digital oligarchy, where I argued that countries in Africa and much of the Global South rely on imported platforms and rented cloud infrastructure with limited influence over how those systems are designed or governed, and that private commercial decisions are increasingly performing a function that looks very much like public policy.

IMPLICATIONS

Adoption does not change the category. Microsoft estimates that 23.1 percent of South Africa’s working-age population used a generative AI product in the first quarter of 2026, the highest rate on the continent, while usage in the Global North grew more than twice as fast as in the Global South. Leading Africa in consumption increases exposure without increasing influence. Three responses follow, none of them heroic. The first is participation in standard-setting, which is the cheapest lever available and the one we have treated as diplomacy rather than as risk management. The second is domestic rules with actual enforcement, since a country that cannot govern deployment within its own borders has forfeited the only authority it definitely possesses. The third, for boards, is exposure management rather than mitigation. A South African bank cannot reduce the probability of power centralisation or dangerous capabilities, but it can map where those risks would reach its business and plan accordingly, which is the same discipline already applied to currency and sovereign risk.

CLOSING TAKEAWAY

There is a particular kind of clarity in being described accurately by people who were not describing you. The study’s authors were mapping a global structure and happened to draw our position within it. What they found is that the parties who can reduce these risks and the parties who will absorb them are not the same parties, and that this misalignment is itself a reason the risks persist. South Africa cannot resolve that from outside. What it can do is stop mistaking adoption for participation. Being the continent’s heaviest user of systems designed elsewhere, governed elsewhere and answerable elsewhere is a position, and it is not the one most of our AI conversation assumes we occupy.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net