I have spent a lot of time thinking about how much more one person could do with better software.

A founder operating a company that would otherwise require twenty people. A doctor whose reach extends beyond the patients physically sitting in front of her. An engineer able to build, deploy and maintain systems that previously required an entire team. A considerable part of my interest in AI, including the evolution of Lucien AI, comes from this idea. I wrote about it in I keep coming back to leverage .

But I keep returning to another version of the same question: how much damage could one person do with that much leverage? What about a company? What about a government that already has more money, infrastructure and military capability than most of the world could assemble between them?

The machinery does not acquire better intentions simply because it becomes more capable. In fact, history shows that scaling capability tends to erode ethics, and for various reasons that I won't go into in this article.

For around two years, I have been wondering whether several things developing around us are pushing humanity toward a crossroads. On one side, we continue operating as competing nations, companies and individuals, with increasingly powerful tools and limited means of restraining one another. On the other, we surrender some of that independence to an authority capable of imposing rules on everyone.

The first could become a route to the annihilation of humankind, either wholly or partly. The second could become something people might recognise in the idea of the Antichrist: an authority that promises (and maybe honestly intends) to save the world and acquires dominion over it in the process.

At this time I don't know whether these are our eventual alternatives. There may be intermediate arrangements between them that work considerably better than either. They may also overlap in ways that make the distinction less useful than it first appears.

These are what I want to think through and document in this post.

Artificial Intelligence as a contemporary example

AI is the example I find hardest to stop thinking about, maybe because I work close enough to it to understand its appeal and dangers.

I am an entrepreneur. I like businesses, financial incentives, useful products and the possibility that building something valuable might make its builder wealthy. I have no interest in pretending financial ambition is inherently disreputable while trying to build companies myself.

But I also think that commercial incentives should have hard limits. A company can capture the financial benefits of a decision while other people absorb its costs – costs that, as we have seen over the last three decades, cuts across socioeconomic, ecological, and political tranches, among others. The greater the distance between the beneficiary and the persons carrying the risk, the less reassuring it becomes to think that market pressure will eventually sort things out.

Suppose a company believes its next AI model will give it an enormous advantage but it also believes further testing would be sensible. Its competitors are preparing releases. Investors expect progress. Customers are waiting. How much delay will it tolerate? How much uncertainty will it interpret as acceptable? We've seen this happen this year with the GPT-5.6, GPT-6 and Claude Fable releases.

Now put that company inside a country whose government firmly believes losing the AI race could mean losing economic and military influence. Caution becomes harder to defend. The competitor is no longer merely another company; it is a rival state whose intentions cannot be trusted. You can see this play out when you consider the US government's relationship and current stance towards Anthropic.

The language is already explicit. The Trump administration's 2025 AI Action Plan frames AI development as a race for global dominance, with economic and military advantages at stake. We are clearly past the point of expecting governments to act as impartial referees in this race. They have their own reasons to win it.

Under those assumptions, even people who would prefer a slower, safer race can find themselves accelerating. Each participant has a reason to keep going. Nobody needs to suddenly wake up wanting to destroy the world.

I cannot categorically tell you that this ends in extinction. I don't even think I am qualified to. Losing jobs, losing political agency and losing control of a sufficiently capable Artificial Intelligence system are different possibilities, with different mechanisms and uncertainties. It would be intellectually lazy to throw them into a bag marked “AI doom” and shake that until an argument falls out, which is something I've noticed happening more often this year when people "whistleblow" on social media.

Still, uncertainty about the final outcome doesn't settle the question of who should be allowed to impose the risk.

Military applications make that question considerably more worthy of contemplating.

In March 2025, NATO acquired Palantir's Maven Smart System. NATO's own description includes intelligence fusion, targeting, battlefield awareness, planning and faster decision-making . These are functions through which a military understands a developing situation and decides what to do about it. Yes, software that helps organise that understanding can matter enormously before anyone fires a weapon; but to what else can that software be applied?

Anduril describes Lattice and its associated systems as providing automated awareness and defense against aerial threats. Shield AI develops software that allows aircraft to operate with disrupted GPS and communications. In one published demonstration involving its Hivemind software , aircraft maneuvered autonomously while the decision to engage remained with a human operator.

These descriptions come from the organizations buying or selling the systems. They establish what is being acquired and promoted; they don't independently establish how reliably it works (for example: reliably constraining an AI system so it can never autonomously decide to engage a target, especially if it has, by itself, designated that target).

They also describe different capabilities. An aircraft flying itself and a system independently choosing whom to kill raise different questions. Calling everything a killer robot would make this easier to write and much less useful to read.

These companies are also connecting their capabilities. Palantir and Shield AI announced a partnership bringing intelligence and command software together with autonomous systems.

Simply put, this is leverage being applied to force. And I must explain at this point that for the purposes of this article I define "force" as "real-world effects" because I believe that these concepts of leverage, force and autonomy aren't limited to military scenarios – they are relevant to business, politics, governance and any other fields where AI is leveraged to scale output.

More information can be processed, more activity coordinated and more machinery operated by fewer people. I understand the attraction, and I am constantly preoccupied by thoughts around achieving similar agentic autonomy in running businesses. However, the risks and consequences change considerably when the action being made easier is a military strike as opposed to a hiring decision.

“A human makes the final decision” sounds reassuring; but as a software engineer who understands what goes into building these AI systems, I would still want to know what making that final decision involves.

Imagine being presented with a ranked list of possible targets, supported by more information than you can independently examine, while everyone around you insists that speed is essential. Assuming you can reject the recommendation, do you have the time, evidence, situational context, and institutional support, to challenge it properly? Would you simply click "Approve" and let the AI carry on? Have we retained the person at the end of the process while steadily removing the conditions under which that person's judgment is useful?

The ICRC identifies related concerns : accelerated decisions, excessive reliance on automated outputs and reduced human control. It also stresses that humans remain legally responsible for decisions involving force.

Responsibility on paper still leaves practical questions. Can investigators reconstruct how a recommendation was produced? Can the people reviewing it see the uncertainty? Can they understand it? Can an operator stop the process without being treated as the obstacle everyone bought the software to eliminate?

There are serious arguments for using these tools. Intercepting an incoming drone could save lives. Better information could help a military avoid civilians. The ICRC acknowledges that carefully designed decision support can assist civilian protection. I would be suspicious of an argument that dismissed those possibilities simply because the technological application was military.

But there are separate decisions here: whether we have identified something accurately, whether attacking it is justified, and whether the campaign it belongs to should be happening at all. Improvements in the first do not settle the others, and the third lies firmly in the domain of the humans using the system. A government could become more effective at conducting a war it should never even have started.

That third point gives me another route to worry about: a system can function as intended but help people exercise power that others have very little ability to resist or regulate. We don't have to wait for AI to develop intentions of its own.

It also complicates the idea that governments will restrain the commercial race. They are buying the products. They are the biggest and dearest paying customers of these AI companies. Military advantage is part of the demand. A company seeking growth and a state seeking superiority can encourage each other, with the progress of an adversary (whether in the form of a rival company or another state) supplying a continuing reason to accelerate. Capitalism helps explain the commercial incentive. Inter-state competition adds another, increasingly more compelling, incentive.

The problem with regulation

There are regulatory efforts. The EU's general-purpose AI obligations began applying in August 2025, with Commission enforcement powers applying from August 2026. It is still too early to treat that enforcement record as a completed experiment, but I have a rough mental picture of how things will play out. The UN has also established an international scientific panel on AI .

These developments matter. They also need to be judged by their actual mandates: a framework for commercial AI or an advisory panel does not automatically provide authority over military uses. Producing an assessment, setting a standard and compelling a powerful actor to comply are very different achievements.

A global regulatory body seems like an obvious answer until I ask what happens when somebody important tells it no.

Can it inspect a program that a government considers a national security secret? Can it force a laboratory to stop or slow down? Who establishes that the danger justifies intervention? Who effects the order?

If the answer depends entirely on the cooperation of the actor being regulated, we have found the boundary of the institution's authority. If it can enforce the decision over that actor's objection, we have begun transferring sovereignty.

If we treat both of these (i.e co-operative regulation vs regulatory sovereignty) as two extremes of a continuum, where in that continuum do we want to settle?

This problem is older than AI

On January 3, 2026, US forces captured Nicolás Maduro in Venezuela and removed him to face criminal charges in the United States. Legal scholars questioned the operation's legality; Reuters also reported their skepticism that international institutions could impose meaningful accountability.

Whatever one thinks of Maduro, that capture raises a question larger than his character. Under what conditions may one state send its military into another and unlawfully abduct its leader? And who can enforce the answer to this question against a state powerful enough to disregard it?

The UN and the International Court of Justice embody attempts to put rules around these relationships but their authority has limits built into its structure. Article 94 of the UN Charter permits recourse to the Security Council when a party fails to comply with an ICJ judgment, but the Security Council's voting arrangements give its permanent members veto power to unilaterally block substantive decisions.

A system that depends on powerful states accepting constraints will struggle when those states consider a constraint intolerable. That observation doesn't make these institutions worthless, but it exemplifies why the existence of an institution and the existence of effective accountability are not interchangeable.

The relationship between private wealth, technology and military power makes this more complicated. In 2025, the US Army announced a Palantir agreement with a ceiling of $10 billion over up to ten years . SpaceX received a national security launch contract with an anticipated value of approximately $5.9 billion .

Peter Thiel, Elon Musk and Donald Trump occupy different positions in this picture. Their beliefs and interests aren't purely identical, but the proximity of companies associated with the first two to the state power exercised by the third makes the concentration of capability difficult to ignore.

To be fair, defense contracts do not, by themselves, establish anything sinister. My concern is what could happen when technical dependence, enormous private wealth, fascist idealism and authoritarian political ambition reinforce one another. Which institution would remain independent enough, and powerful enough, to intervene?

That question makes stronger international authority attractive to contemplate. It also makes the prospect of concentrating even more authority uncomfortable. And this is where my Antichrist concept begins to take root.

The Antichrist concept

The word “Antichrist” needs an explanation here.

I am using it to explore a political possibility, devoid of theological or eschatological definition. Peter Thiel has discussed a closely related dilemma : fear of technological catastrophe could help legitimise an authoritarian world state.

After thinking about this for a few months, I find the underlying concern worth taking seriously, although I am skeptical about the politics surrounding it. Acknowledging that a single global authority could become oppressive does not resolve the danger it was created to address.

This is where the argument becomes difficult for me. It is easy to distrust a distant, all-powerful government, but I don't think this ease should be enough for us to refuse to explore this route.

In Nineteen Eighty-Four, which I read as a young child and intend to read again next month, George Orwell imagines three competing superstates , rather than one unified world government. Oceania, Eurasia and Eastasia remain enemies while sustaining similar systems of domination. And their conflict helps preserve the fear and deprivation on which their rulers depend.

And this presents an intermediary position on the Antichrist-Annihilation continuum: the world can remain politically divided while offering ordinary people very little freedom. Its governments don't need to merge to become totalitarian. And, in fact, Antichrist-Annihilation might not even be a continuum at all. It is possible to have both authoritarian control and catastrophic destruction, as we see in Nineteen Eighty-Four.

And military competition could even help them get there. The same fear of an external enemy that justifies accumulating more power (even with relatively benign intentions, which I think Peter Thiel calls "Pax Americana" in one of his essays – can't recall which exactly) can become a reason to tolerate less dissent at home. We could retain the danger of war while acquiring more of the machinery of domination. These outcomes can coexist.

O'Brien is also unusually candid about the Party's objective: power for its own sake. The novel gives us little reason to believe the rulers are sincerely struggling to balance public welfare against individual liberty. Winston's destruction shows how much they want to possess, including his memory, judgment and private loyalties.

Huxley's Brave New World, another novel I read in childhood and recently finished re-reading, makes the bargain more uncomfortable. Its World State provides stability and manufactured satisfaction through conditioning, consumption and the management of desire. Many inhabitants don't experience their lives as intolerable. They have been made to fit them.

Mustapha Mond also explains that scientific discovery is restricted because it could destabilize the order science helped create. He says: “Science is dangerous; we have to keep it most carefully chained and muzzled.”

This touches a worry I have about regulation: the possibility that protecting an achievement eventually becomes a justification for preventing further achievement. Slowing a dangerous development may be entirely sensible. Permanently treating inquiry as a threat to stability is a different proposition. An institution could slide from one to the other while continuing to use the same reassuring vocabulary.

Survival, prosperity, freedom and advancement are not automatically delivered together. Huxley forces me to ask which of them I mean when I say I want humanity to "do well".

Forster's The Machine Stops adds another possibility. Humanity becomes dependent on a technological system that sustains everyday life, until its failure produces catastrophe. Dependence itself can become a form of rule. People have surrendered capabilities they can no longer recover when they need them. In fact, we can already see a form of this dependence in software engineering: like substance addicts, most software engineers simply cannot perform without their Codex or Claude subscriptions. What happens when startups or governments or the military depends almost completely on AI? And if dependence can be a form of rule, who really rules us when this dependence happens: our governments or the capitalists and companies that make these AI systems?

If dependence can be a form of rule, who really rules us when this dependence happens: our governments or the capitalists and companies that make these AI systems?

These stories aren't evidence that any of these futures will happen, but they are (at least) warnings about what might happen.

Sacrifice individuality for collective good?

There is a distinction I need to make before going further. My earlier thinking about this contrast between Antichrist and Annihilation was that a reasonable compromise could still demand the sacrifice of individuality, even if only in part. But a country's freedom to act and a person's freedom to live are not the same thing.

Restricting a government's ability to wage war could protect individual lives. Restricting a company's ability to impose dangerous experiments on the public could expand the freedom of the people exposed to them. International authority might constrain a state precisely where its citizens cannot.

That possibility deserves room alongside the fear of a world dictatorship.

There are also meaningful arrangements between sovereign states doing whatever they can get away with and a single government deciding everything. Treaties can bind particular activities. Institutions can have limited mandates. A federation can divide powers among levels of government. Countries can share some decisions while retaining others. All of these still leave unanswered my earlier question about enforcement when a sufficiently powerful actor (like a government) decides to treat these constraints as artificial and to ignore them.

We already have evidence that cooperation can achieve extraordinary things. The eradication of smallpox and the Montreal Protocol's reduction of ozone-depleting substances did not require a world state. But these aren't quite the same dilemma as we face now. With smallpox, cooperation clearly meant survival. With AI, companies and governments tend to view cooperation as losing a technological race that could have ramifications across sovereignty and finance and the economy.

Nor does distributing authority automatically eliminate coercion. In Le Guin's The Dispossessed , the anarchist society on Anarres struggles with conformity and informal hierarchy despite its rejection of the state. The absence of a formal ruler doesn't prevent people from acquiring the power to obstruct others. It leaves us with another imperfect arrangement to examine, rather than an easy escape from politics.

And then there is the transition itself. A global government formed through democratic agreement would have a different history and legitimacy from one assembled through conquest or emergency powers. This should raise further questions in our minds. Who agreed? What could they refuse? What happens to dissenters? Can a decision be reversed? Can the people making it be removed?

These questions matter particularly to countries that are unlikely to write the first version of these rules – the so-called developing nations. I would want to know whether “humanity” includes us as participants, or merely as people expected to comply with decisions made elsewhere.

Closing the curtain

I still want the benefits that started this train of thought. I want better tools, more capable people, useful businesses, reliable infrastructure and fewer lives constrained by problems we already have the means to solve. I don't want fear to become an argument for keeping people poor or for preserving every inefficient institution because we understand its failures.

I also find it difficult to accept that whoever can move fastest should acquire the right to decide what everyone else must live with.

I wholeheartedly think that open source AI must be encouraged.

Perhaps increasingly powerful technology will make stronger common authority necessary, even to the point of a single world government. Perhaps a collection of institutions with narrower powers can do the work without becoming a world government. Perhaps neither arrangement will be stable, and we will spend generations adjusting the boundaries between them.

The futures I find most disturbing sit at these extremes of Antichrist vs Annihilation: people surviving within systems they cannot meaningfully question or change or the extinction of humankind. And although these are both scary outcomes, I don't want the dramatic possibility of one to make the dangers of the other invisible.

At the same time, I don't want to assume that every surrender of national independence is a surrender of human freedom. In fact, I don't think so. Maybe some freedoms will become possible only when the powerful are compelled to accept limits. Maybe some freedoms will become possibly only when power is concentrated in one or a few institutions with appropriate representation for ordinary people of different socioeconomic backgrounds.

So, Antichrist, Annihilation or somewhere in-between? That's the question I'll leave you to ponder. I'll write more on this topic as I develop my thoughts over the next few months.