Beyond Turing
There is a threshold whose crossing changes forever the status of the subject: the threshold of the skull. For millennia this frontier remained inaccessible from the outside. It is no longer so. Low-cost electrodes record the cortical activity of someone playing with a headset; portable functional near-infrared spectroscopy reconstructs the perfusion of cerebral areas during athletic training; closed-loop neurofeedback systems modify the user's attentional state through an algorithmic loop that infers, acts, re-learns. The question that arises is not whether this is technically possible - it is - but which intellectual category should govern the passage. This article proposes that the adequate category is neither privacy nor data ownership, but a more ancient one: keeping.
§1. The technical domain of brain-computer interfaces
The term brain-computer interface (BCI) designates any system that acquires signals generated by the central nervous system, decodes them, and translates them into a command or an inference. The currently consolidated taxonomy distinguishes invasive BCIs (cortical microelectrodes, electrocorticography), semi-invasive (subdural arrays) and non-invasive (surface electroencephalography, magnetoencephalography, fNIRS, fMRI). It is the latter family, and EEG consumer-grade devices in particular, that has crossed the consumer-product threshold in recent years: eight-to-fourteen channel devices below one thousand euros now provide signals sufficient to recognise attentional states, cognitive load, mental fatigue, and - with growing accuracy - imagined motor patterns.
On the application side, the domain has fractured into three directions. The first is clinical: post-stroke rehabilitation through motor imagery, prosthetic and orthotic control, ambulatory epileptological monitoring, support for locked-in syndrome. The second is performance-oriented: cognitive state optimisation in precision sports, neurofeedback training for performance anxiety management, combined biofeedback in assisted meditative practices. The third, the youngest and most problematic, is everyday: passive input for extended-reality devices, biometric authentication based on cerebral patterns, affective modulation of interfaces. The major industrial laboratories converge on this latter frontier, and on it has opened the regulatory void that this article seeks to bring into focus.
§2. The threshold: from signal to meaning
A careful distinction must be drawn between what the technology reads and what the technology does not read. A BCI system records an electrical, magnetic, or hemodynamic signal; correlates it statistically with internal states; produces a probabilistic inference. It does not read a thought. The distinction is not pedantic: it is the epistemic fracture between neural correlate and phenomenal content - the distinction John Searle, in the famous Chinese Room argument, formulated as the irreducibility of semantics to pure syntax. A brain decoder, however refined, manipulates symbols. It does not access the meaning the subject attributes to those symbols; at most it predicts that meaning with a certain accuracy, and with a certain noise.
Contemporary philosophy of mind has articulated this irreducibility in different but converging ways. Giulio Tononi's Integrated Information Theory quantifies a Φ (phi) as a measure of a system's informational integration, locating phenomenal consciousness as a property of high-Φ systems - a property that does not transfer through external measurement. Stanislas Dehaene's Global Neuronal Workspace describes a computational architecture of conscious access, but the "access" remains internal. David Chalmers consecrated the residual question as the hard problem: why there is something it is like to be a subject, beyond the functions the subject performs. From distinct theoretical perspectives, all signal an identical residue: what remains when every measurement has been taken.
On this threshold we encounter an ancient legal category: the forum internum, the inner forum of conscience that canon law has always declared inaccessible to any external jurisdiction. The same concept resurfaces in the European Convention on Human Rights as freedom of thought, conscience and religion (art. 9): not a permission, but a sphere by definition removed from the power of those outside it. A BCI, when it aspires to infer states the subject has not externalised, does not infringe a privacy norm: it crosses an ontological threshold.
§3. Neurorights and regulatory gaps
The perception of this problem has generated, since 2017, a doctrinal movement proposing an autonomous category of rights: neurorights. Marcello Ienca and Roberto Andorno, in Life Sciences, Society and Policy (2017), articulated the first nucleus: the right to cognitive liberty, the right to mental privacy, the right to mental integrity, the right to psychological continuity. Rafael Yuste, a Columbia neuroscientist, has promoted the constitutional translation: in 2021 Chile became the first country in the world to constitutionalise the "protection of physical and mental integrity" in explicit connection with neurotechnologies. Spain has included analogous provisions in its Charter of Digital Rights. The OECD published, as early as 2019, a Recommendation on Responsible Innovation in Neurotechnology.
On the European front the picture is more fragmented. Regulation (EU) 2024/1689, known as the AI Act, classifies as high-risk AI systems intended for medical diagnosis and the categories listed in Annex III, but does not establish an autonomous "BCI" category nor specifically discipline consumer devices. Article 5 forbids subliminal manipulative techniques, but with a high evidentiary threshold. The GDPR (Reg. EU 2016/679) protects health-related data as a special category (art. 9), yet the qualification of an EEG signal generated by a consumer device as health data is not self-evident. The Medical Device Regulation (Reg. EU 2017/745) intervenes only on devices intended for clinical use. The Oviedo Convention (1997) protects the integrity of the person but does not contemplate, in its original text, the neurotechnological dimension. ISO/IEC 42001:2023, on AI management systems, is methodologically rich but sectorally neutral. An uncovered area thus remains precisely where applicative pressure is greatest.
§4. The convergence of traditions: keeping
In Greek φυλακή (phulakē), in Latin custodia, in Hebrew שְּמִירָה (shmirah), in Arabic حِفْظ (ḥifẓ): four languages, one single category. Keeping is not property, not control, not surveillance. It is an asymmetric relation in which someone holds for someone else something that does not belong to him, and for which he is accountable. Religious traditions converge in placing the heart - a spiritual, not anatomical category - at the centre of this relation. The book of Proverbs formulates the principle with lapidary density: "Keep your heart with all vigilance; for from it flow the springs of life" (Prov 4:23 - omni custodia serva cor tuum, quia ex ipso vita procedit). Genesis, describing the Adamic vocation, joins two verbs: "to till it and to keep it" (Gen 2:15 - ut operaretur et custodiret). The Catholic tradition roots here the principle of the inner forum, the sanctuary of conscience that neither sovereign nor confessor may expropriate.
The Jewish tradition articulates the same intuition through the Tzelem Elohim (Gen 1:27): man as image of God is exhausted in no external representation of him, and precisely for this reason every measured representation of him is insufficient. The Islamic tradition, in the Qur'an (15:29), declares that Allah has breathed into man of His own rūḥ, the spirit; and classical jurisprudence has developed the concept of ḥifẓ al-ʿaql, the keeping of the intellect, as one of the five maqāṣid, the ultimate ends of the law. The Buddhist tradition, in the principle of anattā, does not deny the subject: it denies that the subject can be "fixed" in a datum. From most distant perspectives, a single convergence: thought is not available to anyone but the thinker himself, and when it does become available it does so as gift, not as capture.
This convergence has a precise engineering consequence, which the closing section makes explicit.
§5. Sub tutela Dei: five design invariants
The expression sub tutela Dei does not function, in an engineering article, as a devotional clause. It functions as an epistemic axis: it names the precedence of an ontological threshold over the technical system approaching that threshold. Recognising this precedence is not surrender; it is better design. Five operational invariants follow from this recognition.
First: computational minimisation at the firmware level. The raw signal must not leave the device unless the function strictly requires it. Transformation into aggregate features takes place on device, according to the edge computing architecture which the course's first thematic nucleus articulates in full.
Second: granular and revocable consent topology. A unique and perpetual consent is, in this domain, a contradiction in terms. Each distinct inference requires a distinct consent act, and each act must be revocable in real time, with retroactive effects on retention.
Third: non-transmissibility of the raw signal to third parties. Raw neural data never constitutes an object of commercial exchange or external training; only derived features, and only within contractually delimited functional perimeters.
Fourth: forensic openness fully accessible to the subject. The audit trail of every inference, every access, every derived algorithmic decision must be consultable by the subject himself without producer mediation.
Fifth, and decisive: function limit. Some inferences the device must never produce - religious orientation, political orientation, sexual orientation, pregnancy status, opinion on identified third persons. Not because of technical impossibility: because of an ontologically grounded design choice. This function limit is the module of residual opacity that makes the system compatible with keeping.
Turing's algorithm halts where keeping begins. The incalculable remainder dwells also at this frontier, perhaps above all at this one: the point at which the machine seems to read thought, and the ontological difference reveals itself in its purest form. Thought does not let itself be captured. It lets itself, possibly, be given. A neurotechnology that honours this difference will not be less powerful: it will be more just, more secure, more faithful to what it claims to serve - the person, and in him that measure no external scale exhausts.