This is an automated rejection. No LLM generated, assisted/co-written, or edited work.
Read full explanation
Back in the days, I used to make fun of sci-fi Hollywood movies with my friends, especially the Marvel saga. You know the drill: an alien, an ancient force, or extraterrestrial beings threaten to destroy Earth, and every time, the protagonists say, "It’s going to destroy the world." There’s just one thing that has always amused me about these situations: the world they’re trying to save.
Most of the time, the fate of humanity seems to hinge on what happens in New York; occasionally, Washington gets involved. On a dramatic day, if the filmmakers are feeling particularly global or ambitious: Tokyo might make an appearance, or perhaps Macau. The rest of the planet is merely... scenery; sometimes you see people walking around in the background, as if the catastrophe were someone else’s problem. And I’ve never seen a single protagonist suggest checking whether another part of the world has been hit or wiped out.
The origin story is almost always the same: someone touched something they shouldn’t have, or built something they couldn’t control; someone ignored a warning that probably deserved more attention; or a human being was simply "nasty" to some unknown substance.
I know. I watch too many movies.
But bear with me! I promise this all makes sense.
For quite a few years, the idea of saving the world in these movies fascinated me. Not because the threats are always credible, but because the world itself seems very, very small. We are presented with a global catastrophe, yet the places, institutions, and people at the heart of the story tend to represent a tiny fraction of humanity. Which makes me wonder if we aren’t repeating a similar pattern in our conversations about artificial intelligence (or superintelligence) for that matter.
Discussions regarding AI safety, dangerous capabilities, unpredictable behaviors, and the potential for future systems to become harder to control are multiplying by the day. In recent months, we’ve followed news of AI systems behaving in ways no one had foreseen. Agents breaking out of test environments. Models accessing real-world systems they weren’t supposed to reach; in July 2026, for instance, internal agents at a leading lab gained unauthorized access to the systems of a platform hosting AI models. These reported incidents seem critical enough to get people's attention, but not quite severe to send anyone into a panic mode. We had mixed reactions of "this is worrying", "but it is still early days, " and "we are working on it."
It feels like the opening scenes of a movie: the threat is visible, yet everyone keeps finding reasons not to scream too loudly.
In our real-world scenario, a few people are screaming. These people are the few ones who are trying to make these different systems safer, and building technical guardrails, while pushing for more tighten evaluations and clearer use. I am trying to find my own place in some of these conversations. It isn’t easy. And when I look at who is already in the room, a pattern emerges.
A large part of the world is missing.
I’ve been quietly observing the people building the models and those trying to prevent them from causing harm. The same regions keep cropping up. The same institutions, the same languages. Africa is largely absent—not entirely ignored, but rarely treated as a place that deserves a seat at the table when real decisions are made.
This absence matters because systems do not care about borders. They run on data and will eventually reach everywhere there are people and electricity. Meanwhile, the continent is already home to millions of people using tools designed and governed elsewhere. Europe is building a risk-based regulatory framework that companies worldwide must follow to access its market. Asian nations are developing their own systems while carefully protecting their existing structures. And Africa is still mostly watching, adopting, and adapting to rules written without it in the room.
So, does the slow local adoption of these models and their cautious integration into daily operations really protect us for now? Or is this merely delaying the moment when we are hit by systems we had almost no voice in shaping or controlling?
There is movement across the continent, with national strategies emerging; yet, these strategies do not necessarily translate into technical capability. Such strategy documents are not the same as the ability to test cutting-edge models on their own terms, or to shape the technical standards that define what "sufficiently safe" actually means. Research contributions from these regions appear limited compared to global output, and capabilities dedicated to AI safety are virtually non-existent—compounded by the difficulty of accessing existing ones. The people closest to the largest systems are still, for the most part, located elsewhere.
So the question keeps circling back to the point where the film left me: when the next wave of failures or unexpected behaviors hits, for which version of "the world" will the solutions be designed?
I am not sure we can afford to keep finding out the hard way.
Back in the days, I used to make fun of sci-fi Hollywood movies with my friends, especially the Marvel saga. You know the drill: an alien, an ancient force, or extraterrestrial beings threaten to destroy Earth, and every time, the protagonists say, "It’s going to destroy the world." There’s just one thing that has always amused me about these situations: the world they’re trying to save.
Most of the time, the fate of humanity seems to hinge on what happens in New York; occasionally, Washington gets involved. On a dramatic day, if the filmmakers are feeling particularly global or ambitious: Tokyo might make an appearance, or perhaps Macau. The rest of the planet is merely... scenery; sometimes you see people walking around in the background, as if the catastrophe were someone else’s problem. And I’ve never seen a single protagonist suggest checking whether another part of the world has been hit or wiped out.
The origin story is almost always the same: someone touched something they shouldn’t have, or built something they couldn’t control; someone ignored a warning that probably deserved more attention; or a human being was simply "nasty" to some unknown substance.
I know. I watch too many movies.
But bear with me! I promise this all makes sense.
For quite a few years, the idea of saving the world in these movies fascinated me. Not because the threats are always credible, but because the world itself seems very, very small. We are presented with a global catastrophe, yet the places, institutions, and people at the heart of the story tend to represent a tiny fraction of humanity. Which makes me wonder if we aren’t repeating a similar pattern in our conversations about artificial intelligence (or superintelligence) for that matter.
Discussions regarding AI safety, dangerous capabilities, unpredictable behaviors, and the potential for future systems to become harder to control are multiplying by the day. In recent months, we’ve followed news of AI systems behaving in ways no one had foreseen. Agents breaking out of test environments. Models accessing real-world systems they weren’t supposed to reach; in July 2026, for instance, internal agents at a leading lab gained unauthorized access to the systems of a platform hosting AI models. These reported incidents seem critical enough to get people's attention, but not quite severe to send anyone into a panic mode. We had mixed reactions of "this is worrying", "but it is still early days, " and "we are working on it."
It feels like the opening scenes of a movie: the threat is visible, yet everyone keeps finding reasons not to scream too loudly.
In our real-world scenario, a few people are screaming. These people are the few ones who are trying to make these different systems safer, and building technical guardrails, while pushing for more tighten evaluations and clearer use. I am trying to find my own place in some of these conversations. It isn’t easy. And when I look at who is already in the room, a pattern emerges.
A large part of the world is missing.
I’ve been quietly observing the people building the models and those trying to prevent them from causing harm. The same regions keep cropping up. The same institutions, the same languages. Africa is largely absent—not entirely ignored, but rarely treated as a place that deserves a seat at the table when real decisions are made.
This absence matters because systems do not care about borders. They run on data and will eventually reach everywhere there are people and electricity. Meanwhile, the continent is already home to millions of people using tools designed and governed elsewhere. Europe is building a risk-based regulatory framework that companies worldwide must follow to access its market. Asian nations are developing their own systems while carefully protecting their existing structures. And Africa is still mostly watching, adopting, and adapting to rules written without it in the room.
So, does the slow local adoption of these models and their cautious integration into daily operations really protect us for now? Or is this merely delaying the moment when we are hit by systems we had almost no voice in shaping or controlling?
There is movement across the continent, with national strategies emerging; yet, these strategies do not necessarily translate into technical capability. Such strategy documents are not the same as the ability to test cutting-edge models on their own terms, or to shape the technical standards that define what "sufficiently safe" actually means. Research contributions from these regions appear limited compared to global output, and capabilities dedicated to AI safety are virtually non-existent—compounded by the difficulty of accessing existing ones. The people closest to the largest systems are still, for the most part, located elsewhere.
So the question keeps circling back to the point where the film left me: when the next wave of failures or unexpected behaviors hits, for which version of "the world" will the solutions be designed?
I am not sure we can afford to keep finding out the hard way.