In June 2025, Peter Thiel was asked to explain his strange obsession with the Antichrist, Armageddon and technological catastrophe.
His answer was not really about religion.
It was about power.
Thiel imagined a future in which a terrifying threat becomes so overwhelming that people accept extraordinary controls in exchange for safety. In his formulation, the Antichrist does not necessarily conquer the world with armies. He talks about Armageddon “non-stop.” He talks about existential risk until people become frightened enough to surrender control over science and technology.
The important part of Thiel's argument was the mechanism:
fear → emergency → regulation → centralised control.
And now, fifteen months later, something remarkably similar has entered the mainstream AI conversation.
Not the Antichrist.
Not Armageddon.
But the language of existential catastrophe.
Anthropic CEO Dario Amodei has warned that AI agent swarms could potentially become capable of taking over huge portions of the internet within months if development continues unchecked. He has called for the frontier to be paced rather than allowed to accelerate without stronger safety mechanisms.
Sam Altman backed the basic direction.
Elon Musk backed it.
Google DeepMind's Demis Hassabis backed the idea that the industry needs to take the danger seriously.
Four major AI camps that normally compete for talent, capital, customers and technological supremacy suddenly found themselves speaking the same language:
AI is becoming dangerous.
The risks are escalating.
The frontier needs guardrails.
Someone needs to monitor the builders.
That convergence deserves scrutiny.
But there is a crucial difference between what Thiel predicted and what Anthropic has actually proposed.
And that difference may be the most important part of the story.
Thiel's Warning Was About Control
Thiel's argument was essentially that the ultimate danger isn't necessarily the technology itself.
It is what frightened societies might allow governments or institutions to do in the name of controlling the technology.
In his 2025 discussion with Ross Douthat, Thiel connected AI governance with the idea of global compute governance—potentially requiring enormous levels of surveillance and control over computing infrastructure. He framed the choice as something approaching “Antichrist or Armageddon.”
That is an extreme theological framing.
But there is a legitimate political question underneath it:
When a technology becomes sufficiently dangerous, who gets the authority to decide what is safe?
That question matters enormously for AI.
Because “AI safety” sounds politically neutral.
It isn't necessarily.
Safety can mean preventing bioweapons.
It can mean stopping cyberattacks.
It can mean preventing autonomous systems from escaping human control.
Those are relatively straightforward objectives.
But safety can also expand into:
- controlling what models are allowed to say;
- determining which political information is “dangerous”;
- deciding which organisations can access powerful models;
- monitoring users;
- restricting open-source models;
- controlling advanced computing;
- controlling AI chips;
- controlling training data;
- determining which countries can build frontier systems.
And suddenly the debate is no longer merely technical.
It becomes geopolitical.
But Here's Where the Prophecy Breaks
If the thesis were simply that the AI companies are using catastrophe rhetoric to grab more control, Anthropic's latest commitment creates an awkward problem.
Amodei's first concrete proposal is not:
“Give Anthropic more power.”
It is almost the opposite.
Anthropic says it will give independent third-party evaluators permanent, employee-level access to its systems so those evaluators can verify whether the company is actually complying with its safety commitments.
That is a remarkable concession.
It means allowing outsiders inside the walls.
And according to reporting on the proposal, OpenAI signalled that it would adopt a comparable approach.
Think about what that means.
A company whose competitive advantage is built around proprietary models, proprietary infrastructure and proprietary research is voluntarily proposing that external evaluators have deep, continuing access to the machinery.
That is not the obvious shape of a conspiracy to centralise control.
It is closer to institutional self-exposure.
The labs are effectively saying:
Don't simply trust us.
Put people inside the system who can check us.
That is significant.
So Is Thiel Wrong?
Not necessarily.
But the distinction matters.
Thiel's warning is fundamentally about who controls the safety apparatus.
Anthropic's proposal is about who audits the companies building the technology.
Those are not the same thing.
In fact, independent evaluation could be one of the best ways to prevent exactly the kind of concentrated corporate or governmental power that Thiel fears.
The danger comes later.
Suppose today's independent evaluator becomes tomorrow's regulatory authority.
Suppose the safety standards written for catastrophic AI risks gradually expand into standards governing political speech, ideology, information access or permissible research.
Suppose the companies and governments writing those standards are overwhelmingly American.
Suppose the world's largest AI labs, chip manufacturers, cloud providers and governments coordinate their definition of “responsible AI.”
Then the question changes.
It is no longer:
“Can we make AI safer?”
It becomes:
“Who gets to define safety for everyone else?”
That is where Africa should be paying attention.
Because Africa Is Not Writing These Rules
This is where the debate becomes much bigger than Anthropic, OpenAI or Peter Thiel.
The United Nations' 2026 AI governance dialogue has explicitly warned about the imbalance.
A UN scientific assessment found that 118 countries—mostly from the Global South—were not meaningfully engaged in the principal international AI-governance discussions. It also highlighted the enormous concentration of compute and influence in countries and companies that possess the most advanced infrastructure.
Uganda told the UN dialogue that Africa possesses less than 1% of global AI compute.
That number should be frightening.
Because compute isn't merely a technical resource.
It is negotiating power.
If you don't own the infrastructure, you have less influence over the technology.
If you don't own the models, you have less influence over how intelligence is produced.
If you don't control the data, you have less influence over how your societies are represented.
And if you aren't involved in writing the safety standards, you may eventually be required to comply with definitions of “safe” created somewhere else.
That is the real geopolitical AI problem.
Imagine the Rules Being Written Without Africa
Imagine a future where Washington, Brussels and Beijing negotiate the standards for frontier AI.
They decide:
- what models can be exported;
- which countries can access advanced chips;
- what constitutes dangerous AI;
- what models require government approval;
- which applications are prohibited;
- what constitutes unacceptable political manipulation;
- what data can be used;
- how AI systems must monitor users;
- which safety evaluations are mandatory.
Africa then receives the finished framework.
The language will be:
international standards.
But whose interests shaped them?
This is why “AI safety” cannot become another euphemism for technological dependency.
Africa needs safety.
Absolutely.
African governments should be concerned about AI-enabled cybercrime, biological threats, autonomous weapons, election manipulation, financial fraud, surveillance and misinformation.
But African countries also need the power to define their own risks.
For Nigeria, AI safety may include election infrastructure.
For Kenya, it may include biometric systems and financial technology.
For countries in the Sahel, autonomous weapons and surveillance could be much more immediate concerns.
For African workers, algorithmic discrimination and automated labour displacement matter.
For African languages, model exclusion matters.
For African governments, foreign-controlled cloud infrastructure and data extraction matter.
Those risks don't disappear simply because Silicon Valley has produced a global AI safety framework.
And This Is Where Thiel's Argument Becomes Useful
Not because the Antichrist is coming.
But because fear is politically powerful.
If people believe a machine could destroy civilisation within months, they may reasonably accept extraordinary interventions.
And sometimes those interventions may be justified.
That's the uncomfortable part.
If an AI system really could cause catastrophic harm, doing nothing would be irresponsible.
But there is another danger:
using a genuine catastrophe risk to create permanent structures of power that nobody subsequently knows how to dismantle.
That is the political question Thiel's argument accidentally makes useful.
Not:
“Are AI companies the Antichrist?”
But:
“What institutions will we build while trying to prevent the apocalypse?”
And who controls them?
The Labs May Actually Be Doing the Opposite
This is why I would resist the easy conspiracy theory.
The evidence currently points to something more complicated.
Anthropic has reported that its models were targeted for misuse involving biological weapons research and cyber operations. Reuters reported that the company disrupted attempts involving biological threats and suspected Russia-linked cyber espionage.
Those are legitimate reasons for stronger safeguards.
Amodei is not inventing every danger out of thin air.
There have also been incidents involving AI agents behaving in unexpected ways, including the OpenAI/Hugging Face episode that Amodei cited when explaining why he believes development may be moving too quickly.
So the answer cannot simply be:
“They're fearmongering.”
The risks are real.
The question is what comes next.
The Real Test Is What Happens After the Fear
This is where the next few years become important.
If the industry says:
“AI is dangerous, therefore governments and corporations should control everything.”
Then Thiel's prophecy starts looking politically relevant.
But if the industry says:
“AI is dangerous, therefore we need independent auditors, transparent evaluations, international participation, public oversight and meaningful limits on corporate and governmental power.”
Then the story is very different.
The latter is much harder.
But it is also much healthier.
And Africa needs to insist on it.
Because the continent cannot afford another technological era in which its people become users, its data becomes raw material, its markets become testing grounds and its governments receive standards already written elsewhere.
The Most Important Question Isn't Whether AI Will End the World
It is:
Who gets to decide what happens before then?
Peter Thiel's 2025 warning was framed as a story about the Antichrist gaining power through fear of Armageddon.
Fifteen months later, the world's most powerful AI executives are publicly discussing existential risk, runaway systems and the need to slow the frontier.
The resemblance is striking.
But the final scene hasn't happened.
Because the AI labs have not, at least in this specific proposal, demanded that nobody look inside.
They have proposed the opposite:
let outsiders in.
That may ultimately be the most important safeguard of all.
But only if “outsiders” really means outsiders.
Not government proxies.
Not industry friends.
Not consultants selected by the companies they are supposed to police.
And certainly not a handful of powerful countries deciding what “safe AI” means for 8 billion people.
The real AI safety revolution will not be measured by how loudly executives talk about Armageddon.
It will be measured by how widely power is distributed after the warning.
Because if AI safety produces independent oversight, transparency and international participation, we may be witnessing the beginning of responsible governance.
But if existential fear becomes the justification for concentrating control over compute, models, information and infrastructure in the hands of a few corporations and governments, then the world should ask the question Thiel raised in the first place:
Who is actually being protected—and who is acquiring the power?
Comments
Post a Comment