The question of who should be allowed to influence or alter a person’s cognitive and emotional states is usually asked as though it were a question about the future. It becomes tractable once we notice how much of it is already settled practice. Parents shape the emotional regulation of their children. Teachers shape attention. Physicians alter mood with drugs that are decades old. Advertisers and platforms shape what people notice, want, and believe at a scale that would have astonished the mid-twentieth century. What neurotechnology changes is narrower and more specific: influence becomes measurable, and it can be delivered by a device a person wears on their head or carries inside their body.
That shift changes the character of the question. Instead of a diffuse cultural negotiation, it becomes a question about permission. Someone has to hold the authority to decide what a device does, to which target, on whose behalf, and by what procedure that decision can be withdrawn. The rest of this essay works through which someone that should be.
Three acts that get called shaping a mind
Public argument runs together three activities that differ in mechanism, evidence, and moral structure.
The first is reading: inferring something about a mental state from neural or behavioral signals. The second is influencing: changing what a person is likely to think, feel, or do without touching the nervous system, through information, incentives, repetition, or social pressure. The third is writing: delivering an intervention that changes neural activity at its source, electrically, magnetically, acoustically, or chemically.
Keeping them apart matters because the strongest scientific results sit in the third category, the most widespread real-world practice sits in the second, and the most alarming scenarios tend to blend all three in combinations that have not been demonstrated at all. A device that turns attempted speech into text is a reading system with an impressive record. A recommender that changes what someone attends to is an influencing system with an ordinary record and a contested one. A closed-loop stimulator that fires when a biomarker crosses a threshold is a writing system that has been tested in humans, in clinics, with implanted hardware, under supervision.
Collapsing the three produces two symmetric errors. It inflates the demonstrated power of hardware by borrowing the ubiquity of influence, and it trivializes influence by demanding that persuasion clear the evidentiary bar that direct neural intervention rightly has to clear.
What has actually been demonstrated
The clearest demonstration of reading comes from speech neuroprosthetics. In 2023, a Stanford-led team reported a speech-to-text brain–computer interface in a participant with amyotrophic lateral sclerosis who could no longer produce intelligible speech (Willett et al., 2023). Drawing on intracortical microelectrode arrays, the system reached a 9.1 percent word error rate on a 50-word vocabulary and 23.8 percent on a 125,000-word vocabulary, decoding at 62 words per minute. That is roughly 3.4 times the previous record for any kind of brain–computer interface, and it begins to approach the pace of ordinary conversation, which runs near 160 words per minute.
The reported details are as instructive as the headline number. The arrays recorded single-neuron spiking. Tuning to speech articulators — jaw, larynx, lips, tongue — was intermixed at the level of individual electrodes rather than laid out in tidy anatomical maps. And the neural representation of articulation persisted years after paralysis, which is what made a trainable decoder possible in the first place.
This is a genuine achievement and a narrow one. It required open-skull surgery, a registered clinical trial, one participant with a specific condition, and a research team that assembled data across many sessions. It shows that the intent to speak leaves a decodable trace in motor cortex. It does not show that anyone can read a person’s thoughts, that the method generalizes across a population, or that consent inside a clinical trial tells us anything about consent inside a workplace.
On the writing side, the demonstrated cases are similarly bounded. Closed-loop deep brain stimulation has been tested in treatment-resistant psychiatric illness, where the device listens to a biomarker and intervenes when a pathological pattern appears rather than delivering a constant dose. These are invasive, clinician-supervised settings, and the results — including a single-participant case whose early report was explicit that one person and one device cannot establish a response rate — are best read as proof that the loop can be closed in humans, not as proof that ordinary inner life can be tuned.
Beyond those boundaries lies speculation, and a great deal of it. Mood adjustment from a consumer headband, real-time lie detection, silent control of a stranger’s decisions, memory editing that installs experiences a person never had: none of these has been demonstrated. Some have partial groundwork in laboratory settings. All of them, if they arrived, would arrive through the same governance questions that the ordinary cases already raise. That is the argument for settling those questions before the applications that make them urgent.
Consent has a structure, not only a form
The easy version of this debate asks whether a person consented. The harder version asks whether the consent meant anything, and the difference is easiest to see in ordinary commerce.
Chile became the first country to litigate the point. Its constitution was amended to protect brain activity and the information derived from it, and in 2023 the Supreme Court ruled in a case brought by Guido Girardi against the neurotechnology company Emotiv. The disagreement was concrete: the complainant argued that the device’s terms of service, the retention of neural data on the company’s own cloud, and the withholding of raw data behind a paid tier left him unable to control information about his own nervous system. Analysts of the ruling describe it as the first judgment anywhere to treat neurodata as a distinct category of personal information, and they also note how poorly existing data protection law fits the problem (Cornejo-Plaza, Cippitani & Pasquino, 2024). Neurodata can be reinterpreted as decoding methods improve, can be re-identified when combined with other records, and can reveal mental activity even when it was collected in what looked like an anonymous form.
Two American states moved earlier and differently. Colorado amended its privacy law to fold “biological data,” explicitly including neural data, into the category of sensitive data requiring opt-in consent (Colorado HB24-1058). California added “neural data” to its definition of sensitive personal information under the consumer privacy framework, defining it as information generated by measuring the activity of the central or peripheral nervous system and excluding data merely inferred from non-neural sources (California SB 1223). Both statutes are consumer privacy instruments. Neither of them decides who may alter a mind; they decide who may collect and keep records about one.
International bodies have gone further and then stopped short of binding anyone. UNESCO’s Recommendation on the Ethics of Neurotechnology, adopted in November 2025, is the first global standard-setting instrument in the field (UNESCO, 2025). It treats the human mind as inviolable in principle, requires strictly voluntary and opt-in use in workplaces, cautions against non-therapeutic use in children whose brains are still developing, and singles out products designed to influence behavior or encourage addiction. It is a recommendation. Member states are invited to align national law; they are not compelled to.
The UN Human Rights Council took a complementary route in April 2025, adopting a resolution that describes neurotechnology as capable of accessing, monitoring, modulating, and altering the nervous system and the mind, and requesting that the Council’s advisory committee draft recommended guidelines for applying the existing human rights framework across the technology’s whole lifecycle (A/HRC/RES/58/6, 2025). Its Special Rapporteur on the right to privacy has pressed the complementary case that neurodata is a specially sensitive category that should be governed by the precautionary principle, with informed consent, rights by design, and prohibitions on coercive use in employment and education (OHCHR, 2025). These are processes being authorized, not rules being imposed, and the difference matters. As of this writing, no international instrument tells a company what it may do to a consenting adult’s brain.
The claimants, and why each has a real interest
If authority has to be assigned, it helps to name the parties and take each one’s claim seriously.
The patient has the strongest presumptive claim and the weakest practical position. Neurotechnology of any consequence is currently medical, which means it arrives attached to a diagnosis, a specialist, a device, and a maintenance schedule. A person who depends on an implanted device to speak or to control tremor has formal consent and very little leverage. Leaving a manufacturer forfeits a function. That is the structure UNESCO is trying to address when it insists that consent be free, explicit, and revocable at any time.
The clinician has the expertise claim and most of the responsibility. The clinician is the party who must decide whether a parameter change is therapeutic or harmful, and often the only one who can explain what a biomarker means. Clinicians also have agendas and blind spots, and the history of medicine includes enthusiasm as well as restraint.
The researcher has the discovery claim, and it is the reason any of this exists. The speech-neuroprosthesis and memory-prosthesis programs succeeded because participants who already had electrodes implanted for clinical reasons agreed to experiments with no personal benefit. A regime that made such research impossible by default would be a regime that kept paralysis permanent out of caution about a hypothetical employer. That outcome carries its own moral weight.
The employer has an interest that should be refused. Productivity monitoring of neural state is exactly the case the UNESCO standard anticipates when it calls for voluntary, opt-in deployment in workplaces. Practical refusal is what makes the refusal meaningful. An employee who must accept neural monitoring to keep a job has not consented; the alternative was unemployment, and the pressure is the mechanism.
The state has the security and public-safety interest, and a long record of abusing it. The imagery of interrogation and courtroom thought-reading is mostly speculative, but the more mundane state interests — benefits eligibility, disability assessment, criminal risk prediction — are not, and none of them should be settled by the technology’s own operators.
The platform or vendor has the commercial claim and the largest information advantage in the room. A vendor knows what the device records, what can be inferred from it, and what the business model requires. The buyer typically knows none of that. Asymmetry of understanding is the central fact of consumer neurotechnology, and any right that ignores it will be a formality.
The parent has the developmental claim, complicated by the fact that the child’s own preferences are precisely what is being shaped. A developing brain is also the brain most responsive to intervention, which makes early use both the most promising case and the least reversible one.
Neurorights: a framework still being argued
One influential response has been to propose new rights rather than to search for them inside existing ones. Ienca and Andorno argued for four: cognitive liberty, mental privacy, mental integrity, and psychological continuity, on the grounds that existing human rights instruments were not written for a world in which mental information could be accessed and manipulated (Ienca & Andorno, 2017). Their case is that the mind was long treated as the last unassailable refuge of the person, and that neuroscience has begun to erode the refuge without any corresponding change in the law that protects it.
The Morningside Group, a cross-disciplinary collection of neuroscientists, clinicians, and ethicists led by Rafael Yuste, framed a parallel set of four concerns that same year: privacy and consent, agency and identity, augmentation, and bias (Yuste et al., 2017). Their paper contains the scenario that has since become canonical: a paralyzed participant in a brain–computer interface trial, frustrated with his research team, watches a robotic hand crush a cup it was handing to an assistant, and cannot say afterwards whether the device malfunctioned or whether his own irritation was translated into motion. It is presented as hypothetical, and its value lies in the question it isolates. When a system sits between intention and action, the ordinary way we assign responsibility becomes ambiguous.
Both papers present neurorights as proposals, and that framing should survive into public argument. Adding a right is not like adding a sensor. It changes what courts must weigh, and it can cut against other interests. Advocates are also not of one mind about whether mental privacy should be absolute; a rule that guaranteed no one could ever infer anything about anyone’s mental state would also forbid the decoding work that lets a person who cannot speak communicate.
Chile’s constitutional amendment and the Girardi judgment show what adoption looks like in practice. A country writes a phrase into its founding document, and then the phrase meets a product, a company, and a contract. The result was narrower than a right and more useful than a slogan: it established that neural data is a category courts must treat with particular care.
Influence without electrodes
There is a case for thinking the largest near-term governance problem has no electrodes in it at all. The mechanisms of influence that already operate at population scale — ranking, recommendation, notification timing, variable reward, the framing of a default — share every structural feature that makes neural intervention concerning, and they lack the surgical threshold that keeps direct intervention rare. They are cheap, deployed continuously, and optimized against measurable engagement.
The UNESCO recommendation reaches in that direction when it calls for regulation of products that may influence behavior or promote addiction and asks for clear, accessible information for consumers. It is the least dramatic clause in the document and probably the most consequential, because it applies to devices already in pockets rather than to implants still in trials.
The argument for treating software influence and neural intervention under a common frame is that both can narrow a person’s range of voluntary action while feeling convenient from the inside. The argument against is that treating them identically would either over-regulate ordinary persuasion or under-regulate hardware, and that the two differ in reversibility, in visibility, and in how directly they act on the nervous system. A defensible position is that they share a criterion — who chose the objective, and can the person inspect and refuse it — while differing in the intensity of scrutiny that criterion requires.
A test that survives contact with real systems
The temptation in these debates is to ask whether a technology is good or bad. The better question is procedural, and it can be applied as a checklist.
Who chose the objective? A controller needs a target, and a target is a value judgment. A system optimized for a patient’s stated goals belongs to a different category from one optimized for productivity, engagement, retention, or institutional compliance, even when the hardware and the firmware are identical.
Can the person inspect what the system is doing? An estimate that is hidden is an authority that cannot be contested. The visible form of this is whether the system reports uncertainty rather than collapsing a guess into a confident label.
Is the effect reversible, and by whom? Reversibility is not only a safety feature. It is what makes consent ongoing rather than a single signature at the start.
Does refusal cost the person something they cannot afford? If the answer is a job, insurance, a school placement, or the ability to speak, the consent is nominal. This is where a rights framework earns its keep, because it converts a personal cost into a legal constraint on whoever imposed it.
Whose capacity grows? A system that restores a function and leaves the person more able in its absence has done something different from one that creates a dependence the person cannot exit. The distinction is easier to state than to measure, and it is the one most likely to be quietly ignored, because dependence is profitable.
Applied to the speech neuroprosthesis, most of the answers come out well, and the array is nonetheless an experiment that only a small number of people can currently access. Applied to workplace neural monitoring, most of the answers come out badly even when every user has clicked a box. The checklist is useful precisely because the underlying disagreements will not be settled in principle. They will be settled case by case, and a checklist is what keeps the reasoning visible.
The case against refusing everything
Restraint has costs, and the costs fall unevenly. A person with locked-in syndrome who regains a way to speak gains something that no abstract interest in mental privacy can offset. A person with treatment-resistant depression whose implanted device fires only when their own brain signals indicate a coming episode receives less stimulation, fewer side effects, and longer device life than a fixed-dose alternative would deliver. These are not marketing claims. They are the substance of the clinical literature, and they are why a neuroright has to be drafted with a duty of access alongside the duty of protection.
There is a real failure mode on that side of the ledger. If the standard for a new intervention becomes proof of no possible misuse, the interventions stall, and the people who would have benefited are the ones who bear the cost of the caution. UNESCO anticipates this by recognizing that failure to pursue beneficial medical applications would itself raise serious ethical questions, and by pressing for equitable access rather than merely for prohibition. Any answer to who may shape a mind has to hold both obligations at once: protect the person who does not want to be touched, and reach the person who needs to be.
Where the argument lands
A workable answer has two layers. The first is a floor that no contract can lower: no non-consensual or coerced modification of cognition or emotion, no neural monitoring as a condition of employment or schooling, no retention of neural data beyond what the person authorized, and no use of inferred mental state to make decisions about a person without their knowledge. These are prohibition-shaped rules, and they are the part of the framework closest to being realized, in UNESCO’s recommendation and in the state privacy statutes that already treat neural data as sensitive.
The second layer is a set of obligations that attach to whoever holds the permission: publish the objective, disclose the mechanism, preserve the ability to refuse and to exit, keep the effect reversible where the technology allows it, and carry the burden of proving benefit rather than assuming it. This layer is deliberately thinner than a right, because the specifics change faster than legislation can follow. What stays constant is the direction of the burden. The party that wants to change a mind should have to explain itself.
None of this requires deciding, in advance, what a mind is or how much of one a machine could emulate. It requires deciding who is allowed to hold the dial, and giving the person whose mind it is a way to reach the dial themselves. The technology will keep arriving. The permission should be granted deliberately, in public, and with an exit that works.
Sources and further reading
- UNESCO, Recommendation on the Ethics of Neurotechnology, adopted 11 November 2025
- UN Human Rights Council, Resolution 58/6, “Neurotechnology and human rights,” 2 April 2025 (A/HRC/RES/58/6)
- OHCHR, “UN expert calls for regulation of neurotechnologies to protect right to privacy,” March 2025
- Colorado General Assembly, HB24-1058, “Protect Privacy of Biological Data,” signed 17 April 2024
- California Legislature, SB 1223, “Consumer privacy: sensitive personal information: neural data,” Chapter 887, 2024
- Cornejo-Plaza, Cippitani & Pasquino, “Chilean Supreme Court ruling on the protection of brain activity: neurorights, personal data protection, and neurodata,” Frontiers in Psychology, 2024
- Ienca & Andorno, “Towards new human rights in the age of neuroscience and neurotechnology,” Life Sciences, Society and Policy, 2017
- Yuste et al., “Four ethical priorities for neurotechnologies and AI,” Nature, 2017
- Willett et al., “A high-performance speech neuroprosthesis,” Nature, 2023
Loading comments…