Grounds for legislating and deciding

The mind is not raw material

The constitutional order of democracies rests on a premise none of them disputes: the person is not a means. From it derive the prohibition of slavery, the inalienability of the body, informed consent in medicine, and the special regime that protects personal data.

Architectures that capture attention invert that premise without declaring it. They turn mental states — curiosity, fear, the need to belong, shame — into inputs of a process whose yield is measured in time spent and in the predictability of behaviour. The person using the service is not what the process is for: they are what it extracts from.

What this opens is not a debate about screens or about habits of consumption. It is the question of whether there is a limit to the economic exploitation of mental life, as there is one for the exploitation of the body, and who is to set it.

Freedom of thought comes before freedom of expression

Article 18 of the International Covenant on Civil and Political Rights protects freedom of thought and conscience. Its inner dimension is absolute: the Human Rights Committee has stated that “no limitations whatsoever” on freedom of thought and conscience are permitted, and Article 4.2 of the Covenant excludes that article from any suspension, even in a public emergency threatening the life of the nation. Freedom of expression, by contrast, does admit restrictions provided by law. The inner forum was protected more strongly than the outer one because it was assumed to be out of reach.

That assumption no longer holds. When the sequence of what a person sees, for how many hours and in what emotional state it leaves them, is set by a system optimised on their own history, the formation of preference ceases to be an entirely private matter. The United Nations Special Rapporteur on freedom of religion or belief devoted his 2021 report on freedom of thought, document A/76/380, to precisely this question.

An uncomfortable consequence follows for the usual debate: whoever defends these architectures by invoking freedom of expression is shielding the right that does admit limits while neglecting the one that does not. There is no freedom of opinion where the opinion is manufactured before it is held.

An electorate is not an audience

The legitimacy of a democratic decision rests on the expressed will being formed rather than induced. The whole edifice — parliamentary debate, the regulated campaign, the day of reflection, limits on government advertising — presupposes citizens whose attention is not allocated in advance by a third party with an interest in the outcome.

Recommender systems allocate attention at population scale, continuously, individually and without public record. False content is not required for the effect to be political: it is enough to decide what is seen, how much of it, and next to what. This is why the content-moderation debate, stalled for a decade, does not reach the problem.

The Digital Services Act already requires more than is commonly acknowledged: explaining the main recommender parameters in plain language and offering options to modify them (Article 27); providing at least one option not based on profiling at very large platforms and search engines (Article 38); assessing and mitigating the systemic risks arising from the design of those systems (Articles 34 and 35); undergoing independent audit (Article 37); and granting data access to vetted researchers (Article 40, implemented by Delegated Regulation 2025/2050, in force since October 2025).

What does not exist is transparency about the outcome. There is no duty to publish aggregate metrics of exposure and reach, to log changes in ranking criteria, or to submit the ranking function to public verification. Article 27 bears on declared parameters, not on the actual distribution of attention. That is where the gap lies, and it is a precise gap: naming it this way, rather than as a general accusation of opacity, is what makes the position defensible.

The mental environment is a matter of sovereignty

A state that cannot inspect the systems determining what its population reads, fears and desires does not exercise complete sovereignty over it, however firm its control of borders and currency. The cognitive infrastructure of almost every country is operated from outside it, under foreign law, and with no duty to answer to its parliament.

This is not a protectionist argument, nor a call to fragment the network. It is the observation of an asymmetry: states accept here a degree of opacity they would not tolerate in energy, banking, medicines or aviation, sectors where the power to inspect is held inseparable from the power to authorise. The difference lies not in the gravity of what is being administered but in the historical moment at which each sector was brought under regulation.

Cognitive sovereignty does not consist in deciding what citizens think. It consists in no private actor being able to do so unchecked. A government that claimed it for itself would commit the very abuse it says it is fighting, and that limit is best stated before an opponent states it.

Most of it is already written

Much of what is claimed as novel is already in force and merely awaits application: the protection of personal data; the best interests of the child under Article 3 of the Convention on the Rights of the Child — “a primary consideration”, in its literal wording; general comment No. 25 of the Committee on the Rights of the Child on children’s rights in relation to the digital environment; the right to health; freedom of thought; and the prohibition of unfair commercial practices.

What is genuinely new is narrower than usually claimed: the direct reading of neural activity by consumer devices, and the inference of mental states from behavioural data. There the law is indeed missing, and there the neurorights movement is concentrated. In 2021 Chile wrote into its Constitution a duty to give particular safeguard to brain activity and to the information derived from it. UNESCO adopted at its November 2025 General Conference the Recommendation on the Ethics of Neurotechnology, which is not binding. Five US states — Colorado and California in 2024, Montana and Connecticut in 2025, Vermont in 2026 — treat neural data as a protected category.

Telling the two planes apart saves time and avoids the commonest error in this field: legislating again what is already legislated and unapplied, while what genuinely lacks a rule stays uncovered.