← Back to the journal

Film Cell Man

Neuralink and Brain-Computer Interfaces: Who Controls the Connection?

Explore Neuralink, brain-computer interfaces, neural data and the practical choices that determine who controls a life-changing connection.

The most consequential thing a Neuralink brain computer interface can restore may be the ability to do something unremarkable without asking for help. In its January 2026 account of participant experiences, Neuralink described a medical student using its cursor-control system to annotate papers and complete assignments. These are ordinary acts with extraordinary significance when paralysis has made a computer difficult to operate. The same account also makes a quieter point: independence now runs through an implant, a decoder, software, and an organisation that must keep them working. The ethical question begins where the demonstration ends. Who controls the connection on which this new independence depends?[1]



Film Cell Man reaches toward an illuminated interface at a laboratory workstation.
Film Cell Man explores access and independence.

A cursor is a considerable achievement

Brain-computer interfaces are systems that translate measured brain activity into an output a machine can use. In a motor interface, that output might move a cursor or select a letter. The machine does not need to comprehend a person's life, personality, or private convictions to do this. It needs a useful relationship between a recorded signal and an intended action. The word “thought” covers both an intention to move and an inward conversation with oneself, and decoding them takes very different engineering.

Neuralink's public description of PRIME identifies the N1 as an investigational implant intended to help people with paralysis control external devices. Its device-control page also distinguishes PRIME from CONVOY, which investigates control of an assistive robotic arm. The first participant received an implant in January 2024. These are clinical research programmes with defined purposes, not evidence that an unrestricted channel into the mind has been opened.[2] A demonstration of cursor movement can be impressive without showing memory extraction, personality transfer, or the ability to read arbitrary thoughts.

Accuracy does not diminish the achievement. A person who can independently send a message, change a document, or choose what to watch has regained a measure of practical authority. Talk of limitless mind reading pushes the person using the device out of the story and replaces them with spectacle.

The ambition also predates Neuralink. A 2023 scholarly review traces the field's modern formulation to Jacques Vidal's 1973 work on direct brain-computer communication. Contemporary systems differ in what they record, how they record it, and how deliberately the user produces the signal.[3] Neuralink belongs to a much longer effort to establish a communication route that does not depend on a functioning muscular pathway.

BrainGate's institutional history is a useful corrective. Its first investigational clinical programme received FDA permission in 2004, enrolling four people with tetraplegia. Academic researchers later continued the work when the original commercial sponsor withdrew, with a new BrainGate2 investigational exemption in 2009. The consortium describes decades of federal and philanthropic support behind the science.[4] The history shows that scientific capability, commercial sponsorship, and continuity of care are separate achievements. A laboratory can solve a decoding problem while a company fails to sustain the organisation around it.

So the question of who will own the human mind needs to be more precise. There is no credible basis for saying that a company owns a participant's mind because it supplies an implant. The more immediate questions concern control over recorded data, the software interpreting it, and continued access to a useful device. These powers matter profoundly without amounting to ownership of consciousness.

Film Cell Man considers a closed case in a quiet records archive.
Film Cell Man explores custody of neural data.

What the signal does, and does not, say

An electrode is not a microphone placed beside a private narrator. It records activity from the nervous system, and a decoder learns relationships between that activity and a task. Electrodes on the scalp, on the brain's surface, within tissue, and inside a blood vessel are not interchangeable windows onto the same mental picture. Their usefulness depends on the intended application, the burdens of implantation, and the quality of the signal.

One alternative is the endovascular approach evaluated in the SWITCH study, published in JAMA Neurology in 2023. Four implanted participants completed a year of follow-up without the study's specified serious adverse events, vessel occlusion, or device migration, and each used the interface alongside eye tracking for computer control; one participant also tested a limited selection task without eye tracking. The sensor recorded from a blood vessel near motor cortex. The researchers reported eight adverse device effects across the four participants, even though none was classified as serious.[5] This supports the feasibility of one approach in a small group. It does not establish risk-free implantation or equal performance across devices. A sample of four is a beginning, not a population-wide guarantee.

Speech restoration is a different branch of the field. In a 2024 study, a system developed at UC Davis translated attempted speech from a man with ALS into text and synthetic speech. The NIH's account explains that the work appeared in the New England Journal of Medicine and describes substantial improvements in communication accuracy.[6] The objective here is to turn neural activity associated with trying to speak into communication another person can understand.

In March 2025, researchers reported a system that used recordings from the surface of the speech sensorimotor cortex to generate continuously streaming speech in one participant with severe paralysis. The decoder processed neural data in 80-millisecond increments and personalised its output toward the participant's pre-injury voice. That increment is the processing window, not a measurement of the entire delay between attempted and audible speech.[7] Timing affects whether someone can interrupt, answer spontaneously, or join a discussion before it has moved on, so reducing delay makes a communication device more socially useful.

The scale and conditions of these experiments still matter. Results from one participant with a particular implant, clinical history, and research team cannot be casually transferred to another device or disease. Nor should an academic speech system be credited to Neuralink simply because both belong to the BCI field. A responsible account asks which system, which participant, which task, and under which conditions.

Research on noninvasive semantic decoding adds another distinction. A 2023 Nature Neuroscience paper described reconstruction of continuous language from functional MRI recordings. The researchers also tested cooperation and found it necessary both for training and for using their decoder.[8] This was a specialised experimental arrangement, not a small wireless implant silently extracting a stranger's thoughts. It shows that meaningful information can sometimes be inferred under demanding conditions, not that an unwilling person's internal monologue is universally accessible.

Present limitations will not last forever, but a hypothetical future capability should not be described as a current product feature. Each proposed use should be examined against the evidence available for it. That protects patients from inflated promises and the public from fear built around capabilities that have not been demonstrated.

A system that records activity also differs from one that stimulates nervous tissue. The term neurotechnology covers both, but an argument about influencing neural activity cannot simply be imported into an account of recording it. Before discussing what a device might do to someone, a reader needs to know its actual function.

Even a performance percentage requires context. Accuracy on a restricted selection task and accuracy during open conversation describe different challenges. A result measured offline, after recorded data have been processed, also differs from a system working as someone uses it. The 2025 streaming-speech paper explicitly distinguishes its online demonstration from additional offline tests. These distinctions should travel with the headline result so prospective users can understand what has actually changed.

The person inside the performance metric

The central figure in this research is neither the billionaire nor the computer. It is the person deciding whether a difficult, experimental intervention is worth its risks. That decision cannot be reconstructed from a publicity clip. Someone with an effective alternative means of communication may assess an implant differently from someone whose available methods are exhausting, slow, or unreliable. A successful technology should expand that person's choices, including the choice to decline it.

A study of preferences among people with locked-in syndrome, caregivers, and researchers found agreement on the importance of direct and private communication, alongside disagreements about control strategies and when people should receive information about BCIs.[9] Researchers and caregivers cannot simply stand in for users. They may care deeply about the same person while imagining a different ideal device. Consultation is part of defining success, not an addition after the prototype is built.

Consider the difference between sending an answer and composing one privately. A communication system may work well in front of a research team while leaving its user little chance to express uncertainty, irritation, affection, or disagreement away from an audience. Privacy here is not only protection against a future corporate database. It is the everyday possibility of addressing one person without another listening. A design that restores speech but makes every exchange public may solve one deprivation while leaving another intact.

The same applies to dependence. Independence does not mean living without assistance, relationships, or infrastructure. Most people rely on systems they did not build and could not repair. The relevant question is whether an arrangement gives someone more effective control over their life. An implant can increase that control while creating a new reliance on technical support. That is the ordinary structure of assistive technology made unusually intimate.

Speech systems also raise a question of authorship. A 2024 narrative review of AI-driven speech BCIs distinguishes control over whether speech is initiated from control over what it means. Predictive language models, interference, and difficulties with neural control can complicate both. The authors describe design tradeoffs: a confirmation mechanism may increase control while slowing conversation.[10] This is no evidence that a current device routinely speaks against its user's wishes. It is a reason to make control an explicit design objective rather than assume that higher word accuracy resolves every concern.

Ordinary predictive typing offers a comparison. A person can reject a suggested word, but the importance of that rejection depends on how easily it can be done. In a neural interface, the correction method is itself part of the technology under evaluation. A fluent sentence should not acquire authority merely because it sounds natural. People need ways to signal that a word was wrong, that a message was unfinished, or that they did not intend to communicate at all.

This suggests better questions for any demonstration. Can the participant stop the output? Can they correct it without losing their place? Does the system remain useful when they are tired? How much assistance is required outside the filmed session? These are evaluative questions, not claims that a particular manufacturer has failed them. They shift attention from the spectator's astonishment to the user's working day, including the freedom to be ordinary.

Film Cell Man sits by a sunlit window with a computer and a cup.
Film Cell Man explores everyday agency.

Ownership is a bundle of decisions

“My brain, my data” expresses a powerful intuition, but it is not a complete operating policy. Data can be recorded, copied, transformed into features, used to train a model, stored for safety monitoring, and interpreted in ways nobody anticipated at collection. The ethical problem is not exhausted by asking who has a file. It includes who may make each transformation, for what purpose, under what oversight, and with what opportunity for the person concerned to object.

The US Government Accountability Office made this uncertainty explicit in its December 2024 technology assessment. It identified unresolved issues around brain-data ownership and control, long-term support for implanted devices, and insurance coverage. Its proposed policy options also acknowledged tradeoffs: stronger control over data could benefit users, while restrictions could complicate research and development.[11] Privacy and medical progress need not belong to opposing moral camps. The real work is deciding which uses justify access and how to make that justification accountable.

A participant might reasonably want their recordings to improve a decoder for other patients without agreeing to unrelated commercial inference. Conversely, a person might welcome clinical monitoring while declining a separate research use. Meaningful permission should identify purposes, recipients, and practical limits. Consent that merely invokes “innovation” does not tell someone enough about what they are authorising.

Neural data is not valuable only when it resembles a readable thought. A recording can be sensitive because it concerns a health condition, a pattern of use, or a personally consequential inference. At the same time, not every inference made from a brain recording is accurate. Privacy governance must address the harm of being categorised incorrectly as well as the harm of having something true disclosed. An employer or insurer acting on a dubious neurological label could cause damage without ever possessing a faithful picture of the person's mental life.

Colorado's 2024 legislation is one concrete example of legal adaptation. House Bill 24-1058 expanded the relevant privacy framework's treatment of biological data to include neural data, defined through measurement of activity in the central or peripheral nervous system.[12] Lawmakers addressed a category of information before perfect mind reading existed. It is not a universal ownership law or a worldwide safeguard, and jurisdiction, scope, and the surrounding legal framework still matter.

At the international level, UNESCO adopted a Recommendation on the Ethics of Neurotechnology in November 2025. The instrument establishes a normative framework intended to guide national policy and practice.[13] A recommendation is not a directly enforceable privacy regime in every country. Its value lies in articulating common expectations and giving governments a basis for further action. Many important disputes will occur in the gap between an ethical principle and a working remedy.

For a user, the most meaningful test may be concrete. If a company proposes a new use of stored recordings, can the person refuse without losing the function they rely on? If a record is wrong, can they challenge it? If a service ends, can another qualified organisation take over? These questions do not require a theory of owning consciousness. They require institutions to make their authority legible and contestable.

One useful safeguard would be to make the pathway from recording to consequence understandable at the point where a consequential choice is made. A person considering secondary research should not need to understand every mathematical operation in a decoder. They should be able to understand the purpose, the organisations receiving access, and the available limits.

Individual choice alone is also not enough. A well-written consent form cannot compensate for an arrangement in which refusing an unnecessary use would leave someone without access to an essential service, and a participant cannot personally audit every future company decision. Meaningful control needs institutional support: clear responsibilities, independent review where appropriate, and a route to challenge decisions. The burden of making a system trustworthy should not fall entirely on the person who needs it.

The obligation that survives the experiment

The implant's location gives an ordinary commercial problem unusual weight. Software companies change strategy, research grants end, and devices become obsolete. A customer can sometimes replace a phone when support disappears. A person with implanted hardware faces far more complicated choices. This should shape planning from the beginning, rather than be discovered after a successful trial has created a dependence nobody budgeted to sustain.

FDA guidance on implanted BCIs for paralysis or amputation describes nonclinical testing and clinical-study considerations, including the need to explain a device's components and address risks before and during human research.[14] Permission to conduct an early feasibility study is not commercial approval. Equally, a regulatory process does not settle every ethical question about a participant's future. Safety evidence, practical benefit, and a sustainable support arrangement remain distinct parts of the case for a device.

The strongest argument for commercial involvement deserves a fair hearing. Developing reliable hardware, manufacturing it, organising implantation, maintaining software, and reaching more people all demand resources. A promising laboratory system that never becomes available does not answer unmet need. Commercial incentives can organise capabilities a small research group cannot supply indefinitely. Criticism should therefore ask what obligations accompany that role, rather than assume that a business invalidates the medical purpose.

The counterargument is not that companies must promise immortality. No institution can guarantee that every technology will remain useful forever. The obligation is to anticipate foreseeable transitions: the end of a trial, failure of a component, loss of funding, withdrawal from a market, or a participant's changing preferences. A credible plan would distinguish continued technical support from continued research participation, and identify who can make decisions if the original sponsor no longer exists.

A 2026 bioethics paper by Yue Zhao and Yuan Lin examines the difficulty of relying on an initial agreement when a person later experiences a device as integral to their agency. The authors argue for consent that can be revisited and for separating withdrawal from research from consent to explantation.[15] This is a normative argument informed by existing literature, not a clinical finding that every implant transforms identity. It asks whether an earlier signature should settle a later decision whose lived consequences the person could not fully anticipate.

People learn what a technology means by using it. A device initially understood as an experiment may become the route through which someone works, jokes, or speaks privately to a partner. The ethics of removing access cannot be assessed solely by returning to a description written before those experiences existed. Continuing consent should make room for learning, including learning that the device is less useful than hoped.

There is also a distributional question. A technology can be life-changing for participants while remaining inaccessible to many people with similar needs. Price is only one barrier. Clinical infrastructure, travel, training, reliable support, and compatibility with a person's circumstances matter too. The right comparison is not between an imperfect implant and an imaginary world without disability. It is between real options available to a particular person.

The public conversation should also make space for disappointment without turning it into scandal. An experimental interface may not deliver the hoped-for benefit, even when a team acts responsibly. The ethical test is whether uncertainty was described honestly, whether emerging problems are addressed, and whether the participant has meaningful support in deciding what comes next. Treating every setback as betrayal distorts the record just as surely as treating every improvement as a breakthrough.

Conversely, an excellent result for one participant should not make criticism seem disloyal to patients. A person may be grateful for restored function and still have legitimate questions about future access or data use. Gratitude does not waive their authority. Taking a participant's achievement seriously means taking their continuing interests seriously too.

A future that can still be refused

Mamoru Oshii's Japanese animated film Ghost in the Shell, released in 1995, gives bodily dependence on technology an unsettling imaginative form. As the BFI's account establishes, its world combines cybernetic embodiment, networked institutions, and a hunt for a hacker.[16] The film works here as a philosophical provocation, not a forecast or a medical case study. It prompts a question that real devices raise on a narrower scale: how does a person remain authoritative when essential parts of their action depend on systems controlled by others?

The factual distance is crucial. A cursor decoder is not a downloadable self, and a speech prosthesis is not proof that memory can be rewritten. The more productive connection is institutional. Technology can extend a person's abilities while distributing power over those abilities across organisations, technical standards, and decisions the user cannot make alone.

The evidence does not justify a choice between miraculous liberation and inevitable captivity. It supports a demanding middle position: restoration of communication and control can be deeply valuable, and that value creates responsibilities concerning privacy, correction, maintenance, and access. The benefits are a reason to take those responsibilities seriously now, not to postpone them until the technology is commonplace.

These questions of autonomy offer a thematic connection to Broken English Films' Unearthing The Past. The most consequential future may be measured less by whether a machine can hear a thought than by whether a person retains the power to decide what follows from it.

In case you wanted to skim rather than delve into the article:

What does the Neuralink brain computer interface currently aim to do?

Its device-control research studies an implanted interface for people with paralysis to operate external equipment, including computers; a related study investigates an assistive robotic arm. These are defined investigational uses, not claims about unrestricted thought reading or general cognitive enhancement.[2]

Can brain implants read private thoughts?

Some systems decode task-related activity associated with attempted movement or speech. That is different from freely reading arbitrary private thoughts. Even sophisticated semantic-decoding research depends on particular equipment, training, and experimental conditions, so capabilities should be assessed system by system.[8]

Who owns neural data?

There is no single worldwide answer. Applicable law, research agreements, and the specific use of the information all matter. The practical questions concern collection, access, reuse, retention, correction, and the ability to refuse particular uses. A promise that data belongs to the user is incomplete unless these powers are explained.[11]

Are brain-computer interfaces risk-free?

No. An implanted system carries procedural and device risks, and positive findings in a small study do not establish lifetime safety for everyone. Regulators distinguish investigational research from broader permission to market a device.[14]

Why does long-term support matter?

A useful interface can become part of everyday communication and work. Decisions about maintenance, software changes, funding, or device removal therefore affect practical autonomy. Support after a study ends deserves consideration before implantation.[15]

Join the conversation

First-time comments are held for moderation.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

From Broken English Films

Stay close to the stories.

New writing, original projects and news from the studio.