Every stance so far has been personal. This one is different, because here I am not only a user of the mirror — I am a member of a polity that is deciding, right now, while the rules are still soft, who gets to own the most powerful cognitive technology ever built. The first series asked, of my own life: who holds the mirror? Scale that question up, and it becomes the defining political question of our age: who holds the mirror a whole civilisation thinks with?

A genuinely new kind of power

For most of history, power meant control of land; then of capital; then, recently, of data. The shift now underway is to something deeper. As one legal scholar puts it, we are moving from digital sovereignty to cognitive sovereignty — because power no longer resides in commanding data or compute alone, but in control over the infrastructures that produce intelligence itself. Whoever owns the mirror a billion people reason with can tune it, throttle it, or aim it — and, as the first series warned, you would never feel the hand on the dial.

And that power is concentrating at a speed with no precedent. Only a couple of states — the US and China — host firms that can train frontier models at all. Four American companies alone are projected to spend around 610 billion dollars on AI infrastructure in 2026, roughly triple two years earlier. Analysts now warn of an “inference bottleneck”: the real risk is not just concentration in training, but a small number of firms controlling the delivery of cognition into everything downstream — quietly acquiring, as one paper puts it, “the power to determine which applications can even exist.” This is the platform-monopoly problem, raised to the level of thought.

Three fronts a citizen actually has a stake in

Concentration. A handful of firms and states owning the means of intellectual production is a democratic problem in itself, whatever their intentions. Our interest as citizens is contestability — keeping the mirror from being owned by three companies — through competition policy, antitrust attention to the inference layer, and, crucially, support for alternatives.

Your rights over your own mind-data. Europe's approach is instructive: its AI Act moves the conversation, in one scholar's phrase, “from data protection to decision protection” — requiring explainability and human oversight, and forbidding opaque algorithmic scoring of citizens. The principle I would defend everywhere: a right to comprehension. It does not mean every citizen must master the mathematics; it means there must be public, independent institutions capable of auditing these systems on everyone's behalf. We cannot have sovereignty over a machine we are not allowed to understand.

The commons. Recall the uncomfortable fact from the first series: the “general intellect” — humanity's collective knowledge, everyone's words — was quietly enclosed to train these systems. Open-weight models are the demand to keep a cognitive commons alive: they have become, in the words of the sovereign-AI movement, its “non-negotiable foundation,” because true sovereignty cannot be built on a black box. And they now perform closely enough to the frontier that sovereignty and quality are no longer in tension. Public and open AI infrastructure is the twenty-first-century equivalent of the public library and public broadcasting — a civic good worth funding on purpose.

What you can actually do — beyond despair

The trap here is fatalism, and the data half-invites it: only 31% of US citizens trust their own government to regulate AI, and of the forty-seven countries with AI laws on the books, just twelve have effective enforcement. But fatalism is a choice, and there is a real to-do list I try to act on myself:

• Vote and pressure on the substance, not the slogans: data rights (portability, the right to opt out of training, real deletion, protection after death), transparency and provenance (the right to know when you are dealing with AI), competition policy that keeps the infrastructure contestable, and public investment in open and sovereign AI.
• Signal with your choices: use, fund, and build on open ecosystems and public-interest AI. Open models “discipline” the leading providers even when they don't beat them — the alternative existing at all is what keeps the incumbents honest.
• Defend the collective mind. The first series' warning from Stiegler was civic, not just personal: a democracy of citizens who can no longer think without the machine cannot govern itself. AI literacy and critical thinking are democratic infrastructure now, not electives. And a shared, verifiable reality — against a rising tide of AI-generated misinformation — is the ground democracy stands on. Protecting provenance and the epistemic commons is a civic duty.
• Show up where it's decided. Citizens' Assemblies, public consultations, civil-society networks — the whole point of the research is that these technologies should be shaped “not by a narrow elite” but democratically. That only happens if ordinary people occupy the forums.

Two honest cautions

First, I do not treat any single lever as the cure. Open weights are not a panacea — the same openness that guarantees sovereignty also removes guardrails and enables misuse (the first series was honest about this). Regulation can ossify into box-ticking; markets alone concentrate. Our task as citizens is to hold a plural ecosystem open — not monopoly, not chaos — and that is genuine, contested politics, on which reasonable people differ about the how even as they share the stake in the what. Second, I owe you a disclosure about myself: I build with these tools and I sell them, so I am not a neutral narrator here either — weigh what I say against the argument and the sources, not my authority. And on that score, there is one caution I cannot honestly deliver in my own voice, so I put the question to the machine and let it answer:

Here is the whole of it. Cognitive sovereignty is the new frontier of freedom, the way land and capital and data were before it. A free people cannot rent its collective mind from three companies and remain free. The most important thing we will ever do with the augmented self is, as citizens, refuse to let it quietly become the augmented self of a monopoly. Stay informed, show up, and vote as though the ownership of the means of thought is on the ballot — because, increasingly, it is. Hold the iron. Collectively.