liberlume-content-v3
lang: en
title: Is a ban, in the world of AI, inevitable?
summary: «Is a ban in AI inevitable?» is the wrong question: not because the answer is no, but because it hides the only ones that matter, which ban, decided by whom and at what cost, including the cost few put on the books. Really, "the people are minors"?
btc-anchor: 956258,00000000000000000001d471dcd8d38e7af855256b48d755af943cf6e426b0d0
prev: sha256:d1958d52b54c418f3cca0283153e583c4846346de957ceaf8dbe40e10c2a12e5
--- body ---
### «Is a ban in AI inevitable?» is the wrong question: not because the answer is no, but because it hides the only ones that matter, which ban, decided by whom and at what cost, including the cost few put on the books. Really, "the people are minors"?

## A question that looks big

"Is a ban, in the world of AI, inevitable?" is one of those questions that
circulate in debates and sound profound. Underneath, almost always, there is a
precise image: the slide toward a dystopian world, in which AI is used only with
someone's permission, capabilities are granted and revoked, and control has won
out over freedom. The real question, at bottom, is this: are we condemning
ourselves to that world? And yet, put this way, it leads nowhere, because
whatever answer you give it, yes or no, it stays useless. It is worth explaining
why, and doing so on precise ground: not the ban that comes from geopolitics (a
model made unavailable by a decision taken elsewhere, as [has already
happened](/en/01-articolo/)), but the ban that comes from a rule. The regulatory
kind, of which the AI Act is the most discussed example. It is a convenient
distinction to begin with; later it will turn out to be less clear-cut than that.

## A ban does not happen, it is decided

Asking whether a ban is "inevitable" treats it the way one treats the weather:
something that either comes or doesn't, but in any case happens to us anyway. For a rule
that is false. A regulatory ban is not a natural event, it is a deliberate act: a
line written by someone, in a law, after someone decided where to draw it.

It is the first misunderstanding the word "inevitable" carries with it: it makes
the author of the decision disappear. It suggests that no one took it, that it
just happened. But a boundary drawn by law always has a hand that draws it, and
forgetting this is the quickest way to stop asking whether it is drawn well.

And it holds all the more for the worst-case scenario. A dystopian world does not
happen the way rain happens: it is built, one rule at a time, one renunciation at
a time. Calling it inevitable is already the first step toward letting it be
built.

## "Yes, in some form it comes." So what?

There is a weak sense in which the answer is obviously yes. Every powerful
technology sooner or later meets rules, and some of those rules are bans: it
happened with drugs, with cars, with finance. In this generic sense, yes, some
ban on AI will certainly come.

But it is an empty answer. It says nothing one can use. Knowing that "some rule
will come" is like knowing that "one day it will rain": true, and useless.
Everything interesting lies in the details that yes erases. And there are three.

## The three distinct questions that single question hides

**Which.** A ban is a line, and everything lies in where it is drawn. "Banning
AI" means nothing; "banning remote biometric identification, in real time, in
public spaces" means something precise, debatable, bounded. Confusing a specific
ban with a general one is exactly what feeds both the panic ("they're banning
everything") and the indifference ("they always ban something anyway"). The
question "is it inevitable?" keeps the line invisible; the only useful question
brings it into focus.

**By whom.** The line is drawn by someone, with their interests and their blind
spots, and it matters who sits at the table when it is drawn. A poorly written
ban can hit those who are small and leave those who are large untouched: the
costs of compliance are absorbed by those with broad shoulders, not by those just
starting out. So a ban meant to limit the power of a few can end up consolidating
it, and make even stronger that dependence on the big providers that people worry
about in words. "Inevitable" says nothing of all this. "Decided by whom" does.

**At what cost, including the one you don't see.** Every ban has a price on both
sides. On one side the harm it prevents, and that is the side that shows and is
claimed. On the other the value it silently forecloses: what can no longer be
done, tried, built. It is the same asymmetry as [governing AI out of caution
alone](/en/caution-in-the-wrong-place/): the cost of banning can be displayed,
the cost of what was never done appears on no balance sheet and no one claims it.
A ban is almost always judged on the harm it prevents, almost never on what it
quietly switches off.

## Banning a use, banning a tool

There is a distinction the first question should always contain, because two bans
that go by the same name can be profoundly different things. Some hit a **use**:
classifying people, deciding their selection, regulating their access to
resources. Others hit a **tool**, or access to a tool: a model judged too capable
to be in just anyone's hands. Under the same formula, "banning something in AI,"
there are two distinct lines of reasoning.

The first arises from a value judgment: there are uses that harm a person's
dignity, illegitimate regardless of how powerful the machine performing them is.
The line tends to stay still, it holds for a stupid model as for a brilliant one,
and it looks at who applies that model more than at who owns it. The second
arises from a risk judgment: this capability, in the wrong hands, does too much
damage. Here there is no single illegitimate use to isolate, there is a power to
contain; and containing a power, more than banning an act, means deciding who is
granted access.

Neither of the two is the "right" ban and the other the wrong one: both have
serious reasons and faults of their own, and this is where it pays to keep
everything on the table. The ban on a use is more precise and coexists with a
world of many tools, but it is still written by someone and can be badly drawn,
too broad, to the point of switching off legitimate uses. The ban on a tool
answers risks that are sometimes real and large, but it has an elusive object
(capability has no sharp boundaries, and it grows), a line that shifts with the
frontier, and a drift hard to ignore: to work it tends to concentrate access in
few hands, that is, toward the very permission-world one would want to avoid.

The two species, side by side:

|               | Banning a **use**          | Banning a **tool**           |
|---------------|----------------------------|------------------------------|
| What it hits  | the act on a person        | access to the model          |
| How it moves  | still, at every capability | mobile, follows the frontier |
| It tends to   | regulate who uses it       | control who accesses         |

From here no clear-cut position emerges, and rightly so. A finer question
emerges. When one asks "which ban," it is not enough to look at *where* the line
passes: what matters is *on what*, on a use or on access to the tool, because the
two forms cost different things and put different things at risk. Treating
"banning AI" as a single block is the quickest way to see neither the reasons for
the one nor the price of the other.

## The false either/or: either we control it or it destroys

Under the question "is it inevitable?" lies a premise rarely said out loud: that
AI is so powerful as to leave only two exits, either you put it under control, or
it becomes destructive. But that "either/or" is constructed, not found. It is a
third way of putting the question badly, and it is the one that makes the ban
look like the only possible road. It is worth dismantling.

First, and in part this has already been seen: "control" is not one single thing.
Control by whom? The form usually proposed, a few actors (states or big
companies) acting as gatekeepers and deciding who can use what, is exactly the
permission-world one wanted to avoid. The concentration of power does not prevent
the dystopian scenario: it is its form. Here the remedy and the disease risk
coinciding.

Second: there is a safety that does not pass through the gate. A technology can
be made less destructive also by how it is made and distributed, not only by what
is forbidden to it: many actors instead of a monopoly, transparency, models that
can be inspected and run on one's own, liability for harm. It is not the safety of
whoever holds the switch, it is that of whoever can see, verify, correct. A safety
by distribution, not by gate. On one condition, though: that the eyes really look.
"Many who verify" is a promised safety, not an automatic one, and [openness is
invoked more often than it is used](/en/caution-in-the-wrong-place/). Inspectable
models that no one inspects offer a paper guarantee.

Third, in fairness: none of this means "no rules, it will be fine." Some risks are
real, and trusting that it has gone smoothly so far is an error of perspective. It
means that between letting it run and banning it there is an enormous space, and
that is where the different evolution lives. The question "control or catastrophe"
erases that space, exactly as "inevitable?" erases the "which, who, at what cost."

## The hard case

That objection, at bottom, we have already brushed against: trusting that it has
gone smoothly so far is an error. But there is a stronger version of it, and it
must be looked in the face. What if this time it really were different? The
quality, quantity and speed of the changes opening up have few precedents, and
the historical analogy with drugs, cars, finance reassures only as long as the
phenomenon is of the same order. Perhaps it is not.

If so, one thing cracks: "correct afterward" presupposes that an afterward comes,
and in time. With rapid and partly irreversible transformations the correction may
come late, and that is a real cost, one that weighs also on the idea of a safety
entrusted to plurality and diffuse control.

And yet none of this turns "is it inevitable?" into a good question. If anything
it makes it worse. If the stakes are higher, higher is the cost of getting the
line wrong on both sides: graver a ban too many, graver a ban that is missing. And
high stakes and high speed together are exactly the condition in which
"inevitable" becomes more seductive, because it is when one stops deliberating and
relies on "there was no choice." Novelty does not dispense with the three
questions, it makes them urgent.

A boundary remains not to be crossed, and it is the one on which everything turns.
"It is a new and big thing" justifies "doing something with care," not
"concentrating access in few hands." Gravity is a reason for caution, not, on its
own, a reason for the gate. Swapping "this is serious" for "so we decide who may"
is the leap the hard case makes easier, and dearer. Here too, all the more here,
the only thing that helps is to hold the three questions firm, and ask them
louder.

## At what level, in whose interest

There is then a reason to ban that has nothing to do with gravity, and that
"safety" hides even better: interest. A frontier model is not just a product, it
is a strategic asset, and that word ends up covering different things, protecting
people, staying ahead of a competitor, denying a capability to an adversary. From
outside, faced with a "safety" restriction, it is almost impossible to say which
of the three is at work. Not too quietly, inside that word, a conflict can also
play out.

A concrete way to see it is to treat AI the way oil or energy are treated: a
resource whose control, more than its use, gives power, and which is granted,
rationed, denied, with licenses and embargoes. Except the analogy breaks just as
it seduces, because a model, unlike a barrel, is copied. It is the commodity that
does not stay in the barrel. What truly is scarce, and truly is controlled like
oil, lies further upstream: the compute and the energy that feed it.

There the scarcity is real, the cutting-edge chips few know how to make and the
machines to produce them effectively come from a single source. It is a detail
that changes the reading: if the scarce resource is the hardware, an open and
efficient model, one that does more with less, works like a lever that defuses
that advantage. Hence a recurring logic: for whoever is ahead on scarcity, closing
models defends a rent; for whoever is behind, opening them erodes it. The same
restriction told as "safety" can then be, seen from this side, also the defense of
a position.

It is worth saying without emphasis: it is not a conspiracy, it is that the
reassuring frame of safety carries a strategic stake with it, and it is honest to
name it. It is also the point where the boundary drawn at the start, between
regulatory ban and geopolitical ban, turns out to be porous: safety is the
category in which the two blur, and [a model made unavailable from
outside](/en/01-articolo/) and a model released by the dropper out of caution are,
at bottom, kin.

The same suspicion holds in reverse. Opening a model is not in itself a noble
gesture: it can be strategy disguised as virtue, and it has a price of its own,
because a powerful capability, once distributed, ends up also in hostile hands.
But this fear too must be handled with suspicion, because "it's too dangerous this
way" is exactly what suits whoever that openness threatens to say: the safety
alarm is convenient on both sides of the gate. No motive, here, is clear, and
neither of the two roads is free.

And here it is best not to delude oneself into reading anyone's mind. Whoever
defends an advantage rarely lies: often they truly believe the ethical version,
because human nature is very good at feeling as a principle what suits it. For
this reason the honest question is not "in good faith?", which no one can answer
even about themselves, but again: what does that ban do, whom does it favor, at
what cost. Intentions are opaque, effects are watched.

## A case already underway

There is no need to imagine it, the permission-world. Right as this is written, a
frontier model among the most awaited does not come out for everyone: it comes out
for a few. The government that asked for its staggered release approves access one
customer at a time, during a preview phase, in the name of the model's
capabilities and of the risks to security, the computer kind above all. It is, to
the letter, the feared scenario: not a clear ban, but a gate where someone
decides, case by case, who may enter.

It is worth looking at it closely, because it shows in action the three things
"inevitable" erases.

It is not a ban, and it is more effective than a ban. No one banned anything, it
was only decided whom to grant. The form is soft, even reasonable, it is only a
preview, it is only caution, and precisely for this it passes without anyone
crying dystopia. The gate is not announced, it is installed.

Whoever accepted does not defend it as a model, they submit to it as a passage:
they let it be known that it is not the preferred road, that for the future a more
sustainable arrangement will be sought. It is the alibi at work. No one truly
claims it, and yet it is done, and the choice disguises itself as an obligatory
step.

And already it is said that it will become the norm, for every frontier model, of
every lab. Here is the "inevitable" as it is born: not yet a law, a practice that
repeats and that, by repeating, begins to look like the natural order of things.
One "inevitable" at a time, as it happens.

One can debate whether caution here is justified, the capabilities at stake are no
joke. But this is exactly the point: the useful question is not "was it
inevitable?", it is which gate, decided by whom, at what cost, including the
silent one of a world that gets used to asking permission.

## "Inevitable" as an alibi

With these three questions lined up, one understands the real trouble with the
word "inevitable": it becomes an alibi, and it works from every side. For whoever
wants to ban, it is the "we had no choice" that shields them from having to defend
the line they drew. For whoever does not feel like thinking about it, it is the
"it happens anyway" that dispenses them from deciding. In both cases it dissolves
the responsibility for a choice that, a choice, it very much is.

It is the same disguise already seen elsewhere: the decision that masks itself as
a non-decision, as fate. Only here it concerns not whoever gives up on adopting,
but whoever sets or submits to a limit. "Inevitable" is the word that lets
everyone put nothing on the record.

## The right question

It is therefore best to drop "is the ban inevitable?" and keep the questions it
hides: which ban, decided by whom, at what cost, including the one few put on the
books. They are not rhetorical questions, they have different answers case by
case, and that is where something is decided.

Regulation is neither the enemy nor a destiny. It is a decision, and like every
decision it deserves to be taken with awareness, weighing the two sides instead of
looking at only one. And another evolution is possible: if safety is sought in
distribution, many who see and verify, instead of in the gate, few who decide for
all. The real risk is not that bans come. It is accepting them, or refusing them,
as if no one had chosen them, and never noticing how much they cost.

The dystopian scenario, if it comes, does not come as a fate: it is built one
"inevitable" at a time. It is this, not the single ban, that is worth putting
attention and energy into, to form an idea of one's own.

Beneath every gate, ultimately, lies the same conviction: that the people are
minors, and that for their own good someone else must decide. Everything comes down
to one question: are the people truly minors, as a line in *Investigation of a
Citizen Above Suspicion* has it, or can we hope they grow up, and learn to decide
for themselves?

## References

- Regulation (EU) 2024/1689 (AI Act), art. 5 (the ban on "real-time" remote
  biometric identification in publicly accessible spaces):
  [eur-lex.europa.eu/eli/reg/2024/1689/oj](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)
- *Investigation of a Citizen Above Suspicion*, directed by Elio Petri, 1970
  ("the people are minors").