Who Will Audit the AI That Africa Buys?
A new industry to vouch for AI safety is forming offshore, and South Africa risks inheriting its verdicts rather than shaping them.
Sign up for my Substack daily AI newsletter here.
See my AI Training course portfolio for corporate Business Leaders here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
A quiet new industry is taking shape around artificial intelligence, and its product is not a chatbot or an agent but trust. As the largest AI laboratories face pressure to prove their systems are safe, a market is forming to vouch for them, made up of evaluators, red-teamers and, before long, accredited auditors. The opening move came this month, when Anthropic hired its first embedded evaluator in a deal worth billions. For South Africa, the temptation is to watch this as a distant corporate story. That would be a mistake. The rules for judging whether AI is safe are being written now, offshore, and a country that adopts these systems without any means to check them is not governing the technology. It is trusting it.
CONTEXT AND BACKGROUND
This month Anthropic named its first embedded evaluator, a team that will sit inside the laboratory to red-team its models and test their safeguards. The evaluator is not a university or a safety charity but Accenture, working through Faculty, the applied-AI firm it acquired in January, and the two companies expect to invest at least a billion dollars each over five years. Accenture is already one of Anthropic’s commercial partners, selling and deploying the very technology it will now judge. Whatever one makes of that, it marks something larger: the birth of an industry whose product is trust, the business of vouching for whether AI is safe enough to deploy.
That industry is still young. The pool of organisations able to rigorously assess a frontier model is thin, and even its architects concede the ecosystem is nascent and needs formal accreditation. Independent evaluators such as METR, a nonprofit that refuses payment for the work, remain rare. The ground rules for what a credible evaluation requires real access, freedom from conflicts, the right to publish, are being drafted now, in a handful of wealthy markets.
INSIGHT AND ANALYSIS
Notice who is absent from that picture. The models, the evaluators and the emerging standards all sit offshore, while South African banks, insurers and government departments adopt the same systems and inherit the same risks. When assurance becomes a product, the question is not only whether it can be trusted, but who is able to buy it, and who is able to judge it.
There is a further danger. As standards harden into accreditation, the cost of compliance can entrench the largest players and shut out those without the resources to meet it. A regime designed in wealthy markets can quietly become a moat that leaves developing economies as rule-takers.
IMPLICATIONS
The answer is not to reject foreign models, which would be neither possible nor wise. It is to build the local capacity to judge them, and to be clear about what judging them here actually means. An evaluation written in London or San Francisco tends to focus on frontier dangers such as cyber or biological misuse. It rarely asks whether a model handles isiZulu, Sesotho or Afrikaans accurately, whether its credit and underwriting decisions are fair under South African conditions, or who is liable when an agent acts under our own contract and consumer law. Those are the questions that matter here, and no offshore certificate answers them.
That is why assurance must be written into the National AI Policy and into public procurement, so that any significant AI system a South African organisation buys comes with independent test evidence relevant to this market, a right to audit, and a named human accountable for it.
This is the essence of cognitive governance. Accountability cannot be delegated to a distant evaluator whose incentives we cannot see. A board that accepts a foreign safety certificate it cannot interrogate has not managed the risk; it has relabelled it. The countries that can verify, and not only consume, will adopt AI on their own terms.
CLOSING TAKEAWAY
The real contest in artificial intelligence is shifting. It is no longer only about who builds the most powerful models, but about who gets to certify they are safe, and on whose terms. That is a market, a form of power, and a question of sovereignty at once. South Africa cannot be a spectator while the referees are appointed elsewhere. If we want AI we can trust, we must be able to check it ourselves. The alternative is to take the world’s most consequential technology on faith, and call that governance.
Author Bio: 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
Comments