WHO GOVERNS THE MACHINE?
AI, Cognitive Sovereignty, and the Future of Democratic Citizenship
Artificial intelligence is still discussed primarily as a technology.
How productive will it be? How many jobs will it replace? How much will GDP grow? Which companies will win? How quickly will more powerful models arrive?
That may be the wrong way to frame the question.
Because a technology capable of intervening in the way we search for information, write, translate, study, work, remember, and interpret reality quickly ceases to be merely a tool.
It becomes cognitive infrastructure.
And once an infrastructure becomes essential, the decisive question is no longer only what it can do.
It is who governs it.
A few weeks ago, in mid-July, Compact pushed to its logical conclusion a position that remains relatively marginal within the Western right. In his essay The AI Apocalypse Is Already Here, Gregory Conti argues that the problem is not merely preventing the catastrophic scenarios associated with some future artificial intelligence. Cultural, political, and anthropological reasons already exist, he claims, to resist the spread of generative AI as we know it today.
It is a radical conclusion.
We are not certain it is ours.
But the question that comes before it is much harder to dismiss.
Does a civilization still exercise sovereignty over itself when it delegates an ever-growing share of the processes through which it understands the world to a handful of privately controlled machines?
The machine speaks
The first battle concerns what appears most harmless.
A question.
We type something into a box and a machine answers.
We have already normalized it.
But the answer does not come from nowhere.
Every model is the product of data, selection, training, instructions, filters, human evaluations, corporate policies, and decisions about what should be encouraged, avoided, explored, or refused.
One does not need a conspiracy theory about “AI bias” to recognize the problem.
Anthropic, one of the leading companies in the field, openly studies the values expressed by its models. In July 2026, it published research showing that Claude displays different value profiles across models and languages, even after controlling for topic, task, and the values expressed by users.
This is inevitable.
There is no perfectly neutral answer to questions about family, authority, identity, freedom, sacrifice, religion, justice, or the common good.
But from this obvious fact emerges a much larger question.
For most of digital history, the battle over freedom of expression has focused on the ability to publish.
Who is allowed to speak?
Who can be removed?
Which content can be censored?
In the age of artificial intelligence, another battle may come before all of these:
who decides what deserves to be found, summarized, remembered, and presented as plausible?
Freedom of speech in the twentieth century was primarily about the right to speak.
Freedom in the twenty-first may also concern the right not to have an invisible machine standing between us and what we are able to know, say, and remember.
This is the first dimension of cognitive sovereignty.
The machine watches
There is another freedom we came to regard as almost natural in the early internet: the ability to separate, at least partially, our identity from our words.
Pseudonyms.
Separate accounts.
Different communities.
A certain possibility of disappearing into the crowd.
That condition may prove far less stable than we imagine.
In February 2026, a group of researchers demonstrated how agents powered by large language models can compare enormous quantities of unstructured text and link pseudonymous identities across different platforms. In the published benchmarks, LLM-based systems reached up to 68 percent recall at 90 percent precision, substantially outperforming the traditional methods examined by the researchers.
Their conclusion is cautious but important: the practical obscurity that has until now protected many pseudonymous users online can no longer be taken for granted.
That does not mean every anonymous account will automatically be identified tomorrow.
It means the cost of surveillance changes dramatically.
What once required hours of human investigation may become an automated process performed at enormous scale.
And that changes the political meaning of anonymity.
A democracy in which every word expressed inside a digital community can easily be linked back to the same person, aggregated with their other opinions, and automatically analyzed is not simply a more efficient version of the previous internet.
It is something different.
The formal freedom to speak may survive.
The material ability to exercise that freedom without pervasive consequences may shrink.
And when the social cost of expression rises, censorship no longer always requires a censor.
The knowledge that one is being watched may be enough.
The machine works
But freedom of expression is only the beginning.
The companies building artificial intelligence do not describe what they are creating as a slightly better version of software.
OpenAI continues to define AGI as highly autonomous systems capable of outperforming humans at most economically valuable work.
Anthropic has said it expects “powerful AI” possibly as soon as late 2026 or early 2027, with intellectual capabilities comparable to or exceeding those of top human experts across many disciplines and the ability to operate autonomously through the digital interfaces used by workers today.
The timelines may be wrong.
The most radical predictions may never materialize.
The history of technology is full of exaggerated promises.
But when the people investing hundreds of billions of dollars into building these systems organize their companies and strategies around the possibility that such a transformation may be close, politics cannot afford to dismiss it as science fiction.
The usual question is:
How many jobs will artificial intelligence replace?
That may not yet be the decisive question.
The more unsettling one is:
What happens to citizenship when a substantial share of citizens is no longer necessary to the economic production of the nation?
After the proletariat
Industrial capitalism produced enormous conflicts.
But even in its most brutal form, a fundamental relationship existed between capital and labor.
The worker could be exploited precisely because he was necessary.
His labor produced something capital wanted.
From that necessity emerged political organization, representation, trade unions, social rights, welfare, and a considerable part of modern democracy.
The citizen of industrial democracies was not merely an abstract bearer of rights.
He worked.
He produced.
He paid taxes.
He consumed.
He created wealth.
He formed families.
In extreme circumstances, he could be called upon to defend the political community itself.
Power needed him.
And that need constituted, at least in part, a form of bargaining power.
But what happens if that relationship changes?
The proletarian could be exploited because his labor was necessary. What happens when the citizen is no longer necessary at all?
This is the most provocative point raised by Conti in Compact. If artificial systems and advanced automation were to drastically reduce the need for human labor, a population could emerge that retains formal political rights while progressively losing its economic function within society.
It would be something different from the proletariat.
A superfluous underclass.
Not necessarily poor in the traditional sense.
It might receive transfers, services, universal income, entertainment, and goods at increasingly low cost.
But it would be structurally dependent on a productive system in which it participates less and less.
And this is where the economic question immediately becomes political.
When power no longer needs the citizen
Liberal democracies developed alongside a particular relationship between the state, the economy, and the population.
Power needed taxpayers.
It needed workers.
It needed soldiers.
It needed producers.
Ultimately, it needed the cooperation of millions of people.
This is not a complete explanation of democracy.
But it is part of its material infrastructure.
Now imagine a society in which an increasing share of economic production depends on computational capital, energy infrastructure, robotics, and artificial intelligence systems controlled by a relatively small number of actors.
The citizen would still, of course, possess a vote.
But what would remain of his material power?
We do not know whether we will reach that point.
And we do not know how quickly.
But even OpenAI, which presents a profoundly optimistic vision of artificial intelligence, now openly acknowledges that the transition toward more advanced systems raises questions about the concentration of power, the distribution of prosperity, and the need for new institutions.
The issue, then, does not belong only to critics of artificial intelligence.
It is already present in the language of those building it.
OpenAI itself speaks openly about the need for policies capable of addressing a transition toward superintelligence that could transform work, knowledge creation, and the distribution of prosperity.
The question is therefore worth asking without resorting to catastrophism:
Can a democracy remain politically egalitarian inside an economy that becomes radically unequal in the distribution of cognitive and productive capacity?
We do not yet have an answer.
But we may have considerably less time than we think to begin searching for one.
The conservative paradox
This is where the debate becomes particularly interesting for the right.
A substantial part of the new American tech right spent years fighting platform censorship, cancel culture, algorithmic arbitrariness, and the concentration of power inside large digital corporations.
Then a technology arrived that may be far more powerful than those platforms.
And part of that same right embraced acceleration.
More models.
More power.
Fewer rules.
More speed.
Competition with China as the geopolitical justification.
Innovation as a national imperative.
There are serious arguments for this position.
A civilization that unilaterally abandons a strategic technology may simply hand it to its adversaries.
But there is also a contradiction that Compact places brutally on the table.
What, precisely, is conservative about radically transforming the human condition through an experimental technology deployed at the greatest possible speed?
“Move fast and break things” may make sense when what risks being broken is a piece of enterprise software.
It becomes considerably harder to apply when the objects of experimentation are:
Human work.
The ability to think.
Social life.
Anonymity.
Trust.
Our relationship with language.
The very structure of citizenship.
Conservatism is rooted, in part, in skepticism toward the idea that everything capable of transformation ought therefore to be transformed.
Prudence is not immobility.
It is the recognition that institutions, habits, and forms of life contain knowledge no planner possesses in full.
A house can change.
It can acquire another room.
A new roof.
A different kitchen.
It can be altered for decades and remain recognizably the same house.
But burning it down and rebuilding it from the ground up is something else entirely.
The conservative point is not that the new is necessarily bad.
It is that not everything we destroy can be rebuilt.
The temptation of refusal
Conti takes this logic much further.
His conclusion is that the properly conservative response to artificial intelligence should not be to accelerate today while hoping to regulate tomorrow.
It should be refusal.
That position is coherent with his argument.
But we are not certain that it solves the problem.
Artificial intelligence has already acquired strategic, scientific, military, and economic dimensions that make it difficult to imagine any single Western country simply halting its development without simultaneously altering the global balance of power.
And AI may produce enormous benefits.
Scientific research.
Medicine.
Productivity.
Access to knowledge.
New capabilities for individuals and small organizations.
The question therefore cannot be reduced to a choice between technological worship and Luddism.
Militant.Media is interested in the question that comes immediately before it.
Who has the authority to say yes?
Cognitive sovereignty
A political community exercises sovereignty when it retains the capacity to make fundamental decisions about its own future.
We have learned to speak of territorial sovereignty.
Energy sovereignty.
Industrial sovereignty.
Digital sovereignty.
Food sovereignty.
Artificial intelligence may force us to add another category.
Cognitive sovereignty.
The capacity of individuals and communities to retain a meaningful degree of control over the infrastructures through which they know, interpret, and represent the world.
It means asking who controls the models.
Who determines their criteria.
Which values are embedded within them.
Which information can be retrieved.
How identity and anonymity are protected.
Where the data resides.
Who owns computational power.
Which jobs are automated.
Who benefits from productivity.
Who bears the costs of transition.
And above all:
who determines the speed of change.
Because time is a political variable.
A society can absorb enormous transformations if it has enough time to build institutions, norms, habits, and new forms of equilibrium.
It is far less clear that it can do so when an industrial revolution is compressed into the space of a few years.
The question, then, is not only whether artificial intelligence is safe.
It is whether we will retain enough power to decide what “safe” should mean.
Who governs the machine?
Perhaps the great political divide of the artificial intelligence age will not map neatly onto the categories of the twentieth century.
Right and left.
Market and state.
Progressives and conservatives.
It may run through all of them.
On one side, those for whom increased technological capability creates, in itself, a sufficient presumption in favor of deployment.
On the other, those who believe there are goods — autonomy, freedom, dignity, community, human agency, cultural continuity — whose value cannot be measured solely through productivity and growth.
We do not yet know where this line will be drawn.
But we know it is beginning to emerge.
Artificial intelligence promises to distribute extraordinary capabilities.
It may also concentrate unprecedented power.
It could free millions of people from repetitive work.
It could create millions of people for whom the economy no longer knows how to provide a meaningful role.
It could democratize knowledge.
It could become the universal filter through which that knowledge is mediated.
It could enormously expand individual freedom.
It could make technically possible forms of surveillance and conformity unimaginable in the age of social media.
These contradictions remain unresolved.
And precisely for that reason, the answer cannot be left exclusively to the engineers who build the models, the investors who finance them, or the regulators who arrive afterward.
A civilization must be able to debate democratically the technology that may transform it before that transformation becomes irreversible.
This may be the essential political question of the age of artificial intelligence.
Not:
How intelligent will the machine become?
But:
Who will govern the machine when the machine begins to govern an ever-growing part of our world?
Further reading
We do not necessarily share Gregory Conti’s conclusion.
But we believe the question he raises deserves to be confronted in full.
If conservatism exists, in part, to preserve the conditions under which human beings can remain free, responsible, and at home in their own society, what should it do when technological progress itself begins to transform those conditions?
Gregory Conti, “The AI Apocalypse Is Already Here,” Compact, July 13, 2026.
https://www.compactmag.com/article/the-ai-apocalypse-is-already-here
MILITANT.MEDIA
Culture is contested.
Stay on the frequency.