In the famous 1637 treatise, Discourse on the Method, French philosopher and mathematician, René Descartes made the famous proclamation, Je pense, donc je suis, translated as ‘I think, therefore I am’. His main argument was that the very act of thinking - and doubting – was the absolute proof of consciousness and existence.
In other words, Descartes was situating the source of our agency as humans in our capacity to reason. What he may not have imagined is that several centuries later, machines would be on the verge of ‘thinking’ better than humans, and that thinking itself would become an object of geopolitical competition.
This June, Artificial Intelligence (AI), the epitome of machine cognition, marks 70 years since it was first delineated as a distinct academic discipline, at the famous Dartmouth Conference in New Hampshire. One of the key organisers of the gathering, John McCarthy, famously stated that the conference was “to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”
Today we know that this is no longer conjecture. The proof of concept is clear: machines can now automate administrative workflows, generate high-quality multimedia, and conduct deep research. No field of study has escaped the disruptive power of AI. Universities and other institutions of higher learning are boxing shadows as they fight to keep control of the gates of knowledge.
Doctors are stretching their imagination at the prospects of conducting multiple AI assisted surgeries across geographies in real time. And by many estimations, 100-person companies will soon be comfortably doing the work of 1000-person companies with each employee sitting on top of a pyramid of highly capable AI agents.
But some key developments in the US this month have quickly shifted focus from the dizzying speed at which AI is evolving, to how its grammar of control is being written. Just two weeks ago, the United States Government issued an unprecedented directive requiring frontier AI systems developer, Anthropic, to suspend access to its most advanced models, Fable 5 and Mythos 5 for foreign nationals, over ‘jailbreak’ concerns. It has since emerged that Mythos, for instance, identified vulnerabilities in highly sensitive and secure US Government computer systems.
The details of the dispute aside, this is an extraordinary assertion of State power. For the first time since the latest wave of AI development began, a government has invoked national sovereign authority to determine access to a frontier AI capability. Subsequent reporting from US media now suggests that in the last 48 hours, the original blanket suspension may be evolving into something more sophisticated. US Commerce Secretary Howard Lutnick is quoted as saying in a letter to Anthropic that, “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model”. This suggests that the US may be moving towards a regime where only a select set of organisations gain access under government-approved conditions, while broader restrictions remain in place.
If that interpretation proves correct, the significance becomes even greater: that governments will now have power to determine who qualifies to access AI systems, under what conditions, and according to which rules. Anthropic initially withdrew the models because it lacked a practical mechanism for verifying nationality at scale. Yet reports that the government is selectively restoring access to approved organisations is intriguing and raises many pertinent questions.
It reminds one of the nuclear governance regime, born out of growing anxieties of the Cold War in the 1960s. The resultant Nuclear Proliferation Treaty (NPT) created a hierarchy between countries that were allowed to possess nuclear weapons and those whose ambitions would be inspected, constrained or discouraged. Critics have charged that the regime creates an ‘apartheid system’ that permanently splits the world into nuclear ‘haves’ and ‘have-nots’, based on an arbitrary cut-off date of January 1, 1967.
However, despite the striking parallels, the nuclear treaty is only an imperfect analogy for AI governance. First, nuclear weapons are weapons of last resort, and a war can easily proceed without them. But AI, while still not indispensable, will soon become the infrastructure of everyday life, woven into how research is conducted, diseases are diagnosed, how firms compete, and even how students learn. To be excluded from it will, therefore, amount to being pushed down the ladder of knowledge production.
The second difference is that the regulation of nuclear technology was never a unilateral affair dictated exclusively by bureaucrats sitting in Washington or London. It needed delicate negotiations that began between the US and the Soviet Union, before spreading to more than 190 countries currently. The unilateral assertion by the US government is therefore instructive and will undoubtedly raise important debates about who ultimately acquires the authority to decide who can use the power that AI is unleashing across sectors globally.
But as if to muddy the waters further, the Anthropic Institute warned just a week ago, that AI systems could soon meaningfully design their own successors and self-improve without human intervention. This prospect, which AI experts call Recursive Self-Improvement (RSI), is perhaps, even consequential than the spat with the US government. Anthropic, the maker of top tier model, Claude, says such a scenario would call for a pause or slowdown in the development of AI technologies to allow regulatory systems and society in general to catch up.
This is a fascinating position to take: on the one hand Anthropic is warning that humans may soon lose control of AI but on the other, it faults the US Government for being too trigger-happy in clamping down on frontier AI models. This is not necessarily a contradiction, but it creates a governance tension that the world will have to confront in the coming months. Because one can argue that governments should promptly – nay, have a duty to - intervene when technology begins to have a mind of its own, but at the same time, if governments invoke this power too readily and arbitrarily, abuse of power risks being normalised.
This is where one must then pause and ask, where is Africa in all this?
To begin with, many African scientists and entrepreneurs are toiling in laboratories, universities and companies driving AI innovation. Yet as governance shifts from open commercial access towards politically mediated trust frameworks, the challenge confronting Africa changes fundamentally. Will African institutions, companies and governments be regarded as trusted participants in the emerging architecture through which access is granted by more powerful actors.
For decades, the continent has participated in new technologies largely as a consumer, not a producer or co-author of the ensuing governance regimes. But in light of these dizzying developments, that dependence may be deepened through a qualification criterion defined by external forces. What criteria will determine who is trusted? Who will review those decisions? Will they be challengeable? And more importantly, will Africa have a say in all that?
Granted, Africans must debate whether the continent will have enough compute, power, and other classes of infrastructure to join the AI party, but the task seems to be evolving from simply avoiding being ‘locked out’, to the even more consequential one of building the scientific, institutional and diplomatic capacity needed to participate credibly in drafting the rules by which intelligence itself will be governed going forward.
Dr Ageyo is the Editor in Chief of the Nation Media Group. He holds a PhD in media studies with a focus on language use in science and environment communication
