Imagine a world where your behaviour can open doors previously beyond your grasp. And with the same speed, where one mistake can exclude you forever. Where your behaviour tells the Machine everything it needs to know about what you can access. The Machine lets you in and locks you out, using the signals you have (unwittingly) generated.
This is what makes technology really hard to argue about; it feels like it is a much more objective arbiter of access than what we have today. However, it is an amplifier. Of how we think and of opportunities and risk. This amplification depends on how it is built and ultimately, who is watching.
On 17 September 2026, Ninety One hosted the “Beyond Alpha” symposium, bringing regulators and the investment industry together to respond to this very challenge. The day wrestled with emerging and transformational frontiers of finance: digital assets, democratisation of private markets and AI. Each of these technologies holds power to open access, provided the design is intentional.
Take data privacy, which occupied some of the conversation. AI draws conclusions from data and therefore it is only as good as what it is fed. It is tempting to conclude that it is in our interests to allow the machines to see and use all data to become more accurate. Of course, this butts up directly against the concept of ‘consent’ as presented by POPIA and its European counterpart, GDPR.
The somewhat fashionable response to this is ‘Privacy Nihilism’. This describes the resignation that digital surveillance is all encompassing and that trying to protect one’s data privacy is a pointless endeavour. In 2018, Ian Bogost coined the phenomenon in his essay in the Atlantic: “Processed foods look wholesome compared to your processed data, scattered to the winds of a thousand databases.”
And this was before generative AI.
Is there even a realistic alternative to surrendering?
Consent, while critical, has less impact in a world of inference. In other words, even if you opt out of every bit of marketing material in the world, you leave a digital footprint that is a treasure trove for conclusions to be drawn. E-commerce used this first. Most of us engage in what the industry terms “cursor drooling”, where we spend longer hovering over items we prefer. Sometimes the difference is milliseconds. This hesitation is ingested by the machine to sell products aligned with your preferences. These preferences are never stated and precisely derived by the algorithms.
Our realities are being optimised for us. What, then, is the alternative to surrender?
Digital autonomy reframes the response from the flat instrument of consent (which asks what you agree to hand over) to something more responsive to outcomes. Digital autonomy considers how the generated data, what the machine assumed about you rather than what you said, is being used. In practice that means three things: that you can see what has been inferred about you, that you can contest it, and that an adverse inference expires rather than following you for life.
In the retail example, the worst-case scenario is a poor advert or irrelevant product.
Of course, the stakes for financial services are much higher. Inference is already built into the industry at cohort level: suitability, risk appetite and grouped behavioural assumptions. AI moves that from the cohort to the individual, at scale. Imagine your digital footprint is used to prove eligibility or personalise savings and lending rates.
These aren’t the daydreams of a sci-fi writer.
In 2020, Bloomberg featured the story of Kenyans who were being charged high interest rates through tech start-ups promising financial inclusion. These lending apps foreshadowed what AI inference can do at scale. Specifically, the application process allowed for personal data like location logs to be accessed. This information was used to credit-rate individuals and allow them credit when they may have been denied access from traditional lending institutions. By the time the article was published, roughly 2.5 million Kenyans had been reported to credit reference bureaus by digital lenders. The promise was inclusion and the result for many was irreversible exclusion triggered by trivial sums.
And yet, many of those borrowers were offered credit precisely because of their behavioural data. Without it, traditional banking channels would never have cleared them.
Interestingly, Kenya’s regulator responded and quickly at that. In April 2020, the Central Bank of Kenya set a Ksh1,000 (about R125) floor for negative credit listings and delisted borrowers who had been blacklisted only for sums below it. It withdrew unregulated digital and credit-only lenders’ access to the credit information sharing system altogether. A licensing regime followed in 2022. By May 2026 the Central Bank had licensed more than 220 digital credit providers, which between them had advanced some 8.4 million loans worth around KSh150bn (roughly R20bn).1
So, Kenya didn’t write off the technology completely, but found a way to harness the benefits (widening the net for access) and blunt the harms (permanent exclusion for de minimis defaults).
Therefore, the generalised legal approach to mobilising the power of the technology could be banning data from being weaponised: make it easy to use data to let someone in, and hard to use it to shut them out. The exclusionary decision should carry the higher evidential bar, a route of appeal, and an expiry date. The traditional finance analogy already exists – it is typically easier to open a bank account than to have one frozen.
These are not easy conversations. But the size of the prize is significant, particularly in a country like South Africa where the formally excluded are not a minority. New technologies could be gamechangers for financial inclusion, provided the design is right. The design determines whether these are door openers or shackles for life. This is not a problem for Silicon Valley to solve on our behalf. It is squarely ours and time is ticking.
1. Central bank of Kenya, Press release, 14 July 2026, 'Licensing of digital credit providers'
Authored by
Featured event
