“Restoring sight to the blind”: what Neuralink’s Blindsight implant can actually do

Elon Musk has announced that Neuralink will implant its Blindsight device in a first blind person within the coming months. The promise is spectacular, and it is currently circulating unchallenged in the French-language press. This article sets out neither to applaud it nor to demolish it: it separates what has been demonstrated from what has been announced, and answers the only question that matters in the consulting room — does this concern me, and should I wait?

THE ANNOUNCEMENT

What has been announced, and what can be verified

Neuralink is developing two distinct devices. Telepathy, the motor implant, allows paralysed people to control a computer by thought: it has so far been implanted in a limited number of patients, within early feasibility studies. Blindsight is a different project, intended to restore a form of visual perception by stimulating the visual cortex directly. It is Blindsight that is meant when people speak of “restoring sight to the blind”.

Three things deserve to be carefully distinguished.

  • What is publicly documented: Neuralink has registered seven protocols on the public clinicaltrials.gov registry (PRIME in the United States, CAN-PRIME in Canada, GB-PRIME in the United Kingdom, UAE-PRIME in the United Arab Emirates under the identifier NCT06992596, VOICE, and two others). All of them carry the N1 implant reference and concern tetraplegia, cervical spinal cord injury, amyotrophic lateral sclerosis or brainstem stroke. None carries an ophthalmological indication.
  • What rests on the company’s word alone: the FDA “Breakthrough Device” designation obtained in September 2024. This point calls for an unusual clarification — the FDA states explicitly that it does not make granted designations public before a marketing authorisation is issued. There is therefore no official registry allowing this information to be verified independently. The source is Neuralink.
  • What belongs to projection: the timeline (“a first implantation within six to twelve months”), the promise of vision “superior to natural vision”, the perception of infrared. None of these rests on a published result.

“Breakthrough Device” does not mean “approved by the FDA”

This is the most widespread confusion, and it is repeated as such by most French-language articles. The Breakthrough Devices programme is defined by US law (21 U.S.C. §360e-3). Its purpose is to speed up exchanges between the manufacturer and the agency, and to prioritise review of the submission. The text contains an explicit clause, at paragraph (g), specifying that nothing in this section alters the criteria and standards of evaluation of an authorisation application.

The order of magnitude speaks for itself: according to the figures published by the FDA as of 31 March 2026, 1,284 “Breakthrough Device” designations have been granted, against 198 marketing authorisations obtained for the designated indication — about 15%. A designation is a starting point, not a validation.

A timeline of the announcements, date by date

Rather than comment on the announcements, the simplest approach is to put them back in order and date them. Each entry below links to its original source. The reading speaks for itself.

  1. 17 September 2024 Neuralink — official announcement, neuralink.com

    “We have received Breakthrough Device Designation from the FDA for Blindsight, for individuals with vision impairment.”

    That is the full text of the announcement. It mentions no clinical trial, no patient, no timeline. The designation opens an accelerated channel of exchange with the FDA: it authorises nothing.

  2. 27 September 2024 IEEE Spectrum, Greg Uyeno — spectrum.ieee.org

    Ten days later, the journal of the Institute of Electrical and Electronics Engineers — the world’s largest professional society in electrical engineering, hardly suspect of technophobia — ran the headline: “Neuralink’s Blindsight Implant Won’t Deliver Natural Sight”. The engineers’ reservations are therefore contemporary with the announcement, not later.

  3. 31 March 2025 Elon Musk, public meeting in Wisconsin — remarks reported by Newsweek, 31 March 2025

    “We’re hoping later this year to do our first device implant for a human, enabling someone who is completely blind to see. It will be low-res at first, so I want to set expectations accordingly.”

    Remarks made in English, quoted here in the speaker’s own words.

    The caveat comes from Musk himself, and it is rarely picked up: low-res, “I want to set expectations accordingly”. It is the second half of the sentence that disappears when the press repeats it.

  4. 28 January 2026 Neuralink — progress report “Two Years of Telepathy”, neuralink.com

    The company published its two-year report: 21 participants implanted worldwide, and the launch of a new trial, VOICE, devoted to restoring speech. The indications cited are spinal cord injury, amyotrophic lateral sclerosis and brainstem stroke.

    This document contains neither the word “Blindsight”, nor “vision”, nor “visual cortex”. Twenty-one patients implanted, none for sight: the information comes from Neuralink, not from its critics. It is the most telling item in this timeline.

  5. 17 August 2026 Public registries — clinicaltrials.gov, PubMed

    The first implantation announced in March 2025 for “later this year” has not taken place. No Blindsight trial is registered on clinicaltrials.gov. No implantation is confirmed. No human data, for any Neuralink device, has been published in a peer-reviewed journal, or deposited as a preprint.

Nothing in this timeline says that Blindsight will not work. It says something more limited, and sufficient for the consulting room: as things stand, there is no result to evaluate. An announced deadline slipping is commonplace in medical research, and is not in itself a criticism. What changes the reading is that no data has come to document it in the meantime.

THE PRINCIPLE

Blindsight does not repair the eye: it bypasses it

This is the point that media coverage most often skips over, and yet it is the one that determines who the technology is for. To produce an artificial visual perception, one can act at three levels of the visual pathway.

  • At the retina (subretinal or epiretinal implant). The implant replaces the destroyed photoreceptors, but relies on the patient’s bipolar cells, ganglion cells and optic nerve, which must be functional. This is the principle of the PRIMA implant and, before it, of the Argus II.
  • At the optic nerve. An avenue that has been explored, but little developed in clinical practice.
  • At the occipital visual cortex. The eye and the optic nerve are set aside entirely, and the brain is stimulated directly. This is the principle of Blindsight, and also that of the academic ICVP device developed in the United States, or of the Spanish work by the Miguel Hernández team.

The consequence is direct. A cortical implant is designed for people whose eyes or optic nerves can no longer transmit anything: bilateral enucleation, complete optic atrophy, certain injuries. For a patient whose retina is diseased but whose optic nerve works, it is not the right technology — and it will be no more so tomorrow, because this is not a question of maturity but of anatomical target.

WHAT IT PRODUCES

What does a person fitted with a cortical implant actually “see”?

The answer has been known for a long time, and it is more modest than the word “see” leads one to imagine. An electrode stimulating the occipital cortex produces a phosphene: a point of light, with no stable colour and no sharp outline, perceived at a location in the visual field that depends on the position of the electrode.

The phenomenon has been described since 1968. Brindley and Lewin implanted an array of electrodes at the occipital pole of a 52-year-old blind patient and reported stable phosphenes, distinguishable from one another as soon as the electrodes were about 2.4 mm apart. They already noted a problem that is still unresolved: phosphenes move with eye movements, which is not the case with a real image.

If you want a sensory idea of what this involves, we have devoted an article to phosphenes, those flashes of light in the visual field that many patients describe spontaneously. A cortical implant produces nothing other than that — but ordered phosphenes, driven by a camera.

The most advanced human result to date: one single patient, six months

In 2021, Eduardo Fernández’s team at the Miguel Hernández University in Elche published in the Journal of Clinical Investigation the most advanced result obtained in humans with an intracortical array. Its exact terms need to be read: one single participant, aged 57, completely blind, fitted with a 96-electrode array implanted in the occipital cortex for six months, then explanted. She was able to identify some letters and to recognise the outlines of simple objects. No visual acuity is reported in this publication.

It is a remarkable demonstration of feasibility. It is not a restoration of sight, and the authors do not claim that it is.

The pixel screen is a poor metaphor — and this has been demonstrated

The natural intuition is to picture the electrodes as the pixels of a screen: the more there are, the finer the image would be. A US team tested this hypothesis and published the result in Cell in 2020. When the electrodes are switched on simultaneously, like pixels, shape recognition is poor. When they are switched on in sequence, to “trace” the shape on the surface of the cortex as one would write with a finger, the participants — sighted and blind — recognise the shapes accurately, up to 86 shapes per minute in the blind participants.

This result bears directly on any assessment of Blindsight: the visual cortex is not a display surface, and the pixel-based principle put forward in the company’s communications is precisely the one this study finds wanting.

THE LIMIT

Why adding electrodes will not be enough

The most advanced result in terms of channel count comes from a study published in Science in 2020: a 1,024-electrode prosthesis implanted in areas V1 and V4, in monkeys. The animals immediately recognised simple shapes, movements and letters. Two reservations must be stressed: these are animals, and they are sighted monkeys, trained — not a cortex deprived of visual input for decades.

As for how many electrodes would be needed for useful vision, the literature offers benchmarks — but they are not comparable with one another, and adding them together would produce a false figure. Each result holds for its own task and its own visual field:

  • The simulation work of Cha and colleagues (1992) obtains an acuity of the order of 20/30 — about 6.5/10e on the French decimal scale — with 625 elements — but this is an extrapolation, in sighted subjects, over a foveal field of 1.7 degrees, through a perforated mask. In the same configuration, reading scrolling text reaches about 170 words per minute.
  • The Geneva work of Sommerhalder and colleagues (2004) achieves the reading of a full page with about 600 contacts — but at 15 degrees of eccentricity, after two months of training at one hour a day, and at a final speed of 14 to 28 words per minute.

Above all, a modelling study published in 2024 in Scientific Reports, built from all the available human cortical stimulation studies, concludes that the perceptual quality of cortical prostheses is likely to be limited by the neurophysiological organisation of the visual cortex itself, rather than by engineering constraints. In other words: the number of electrodes is not the limiting factor one imagines. This is the strongest argument to set against any promise founded on channel count alone.

THE PRECEDENT

What the history of the “bionic eye” has taught ophthalmologists

The Argus II was the first retinal implant to obtain a CE mark (2011) and then a US authorisation (2013). The five-year results, published in Ophthalmology in 2016, cover 30 patients with end-stage retinitis pigmentosa. They are instructive, provided they are read precisely:

  • 24 of the 30 patients still had a working system at five years, and the study’s primary endpoint was safety, not acuity.
  • On the grating acuity test, 38.1% of patients performed significantly better with the system on than with it off at five years. For square localisation, 80.9%.
  • The highest acuity ever recorded across the whole programme is 20/1260 — in a single patient, reported in 2012. It is not an average. In French notation this corresponds to roughly 1/60e: the person distinguishes at 20 feet what a normal eye distinguishes at 1,260 feet.
  • The “letter reading” reported in 2013 involved isolated letters, in a forced-choice task, with a minimum size of 0.9 cm presented at 30 cm — the equivalent of a one-centimetre letter seen at arm’s length, in six subjects.

The comparator here is not “seeing normally”: it is “seeing nothing at all”. On that basis, these results have real value for the patients concerned.

The lesson that matters is not medical, it is industrial

Second Sight, the manufacturer of the Argus II, stopped production of the retinal implant in 2019, came close to disappearing in 2020, then merged with another company whose priority is a drug-delivery implant. According to the investigation published by IEEE Spectrum in 2022, more than 350 blind people worldwide now carry a device that is no longer supported, with no updates and no repairs possible. A non-functioning system inside the eye can interfere with certain imaging examinations, and removing it is neither trivial nor free of charge.

The subject has become a field of study in its own right: a systematic review published in 2024 in JAMA Network Open proposes a consensus definition of the abandonment of an implanted neurological device. This is the question a patient should ask before any other about an implant offered by a private company: who will maintain this device in fifteen years?

WHO IT IS FOR

If you are followed for AMD, diabetes or glaucoma, this is not your technology

This is the sentence found nowhere in the French-language coverage of Blindsight, and it is probably the most useful one.

A cortical implant is intended for people whose visual pathway is completely interrupted upstream of the brain. Yet in the vast majority of situations encountered in an ophthalmology consultation:

  • AMD, even when advanced, leaves peripheral vision and a functioning optic nerve.
  • Diabetic retinopathy damages the retina unevenly, and a large part of the visual pathway remains usable.
  • Glaucoma progressively destroys the fibres of the optic nerve, but the damage is rarely total and, above all, avoidable if it is detected.
  • A cataract, an operated retinal detachment, an epiretinal membrane: the visual pathway is intact or repairable.

In all these cases, the question is not “should I wait for the implant?” but “is the existing treatment being applied at the right moment?”. These are two unrelated problems.

The concrete risk: delaying a treatment that works

This is the only point in this article that bears directly on care. In retinal disease, time is not neutral. Destroyed retina does not regenerate, and a delayed treatment does not recover what has been lost:

  • In exudative AMD, intravitreal anti-VEGF injections aim to stabilise the disease; their benefit depends closely on how early and how regularly the treatment is given.
  • In proliferative diabetic retinopathy, panretinal laser photocoagulation is intended to prevent haemorrhagic complications and tractional retinal detachment. Laser carried out in time prevents a loss that would afterwards be irreversible.
  • A retinal detachment is a surgical emergency: the time to treatment determines the visual outcome.

No implant, cortical or retinal, currently restores vision comparable to what a treatment carried out in time makes it possible to preserve. Postponing an injection, a laser session or surgery while waiting for an announced technology amounts to trading real vision for hypothetical vision.

WHAT IS GENUINELY MOVING FORWARD

Meanwhile, on the retinal side, something has moved

The paradox of this news cycle is that a genuinely documented result went largely unnoticed alongside the Neuralink announcements. The PRIMAvera trial, on a subretinal photovoltaic implant in patients with atrophic AMD, was published in the New England Journal of Medicine.

  • 38 patients implanted, 32 evaluable at twelve months.
  • 81% (26 of 32) show a gain of at least 0.2 logMAR, that is 10 letters or more, on the primary endpoint.
  • Mean gain at twelve months: 25.5 letters. Mean prosthetic acuity: 1.32 logMAR, that is about 20/417 — of the order of 1/20e in French notation.
  • Natural peripheral vision remained unchanged after implantation.
  • 26 serious adverse events in 19 participants, 81% of them within the two months following surgery and 95% resolved within two months. This is an open-label, single-arm study, with neither masking nor randomisation.

Two cautions are called for, and they matter. First, the patients in the PRIMA trial are not blind: they have a central scotoma from atrophic AMD and retain their peripheral vision. Confusing that situation with the one Blindsight targets would be a complete error of scale. Second, an open-label single-arm study does not support the same conclusions as a randomised trial. We have set out this file in a dedicated article: the PRIMA implant and what it changes in AMD.

Other avenues are also moving forward, notably retinal tissue transplantation. None of them is available in routine practice in France today.

CONGENITAL BLINDNESS

What about people blind from birth?

This is the most fragile claim among those in circulation. A visual cortex deprived of input since birth does not remain on standby: it is partly recruited by other sensory modalities. A study published in Nature in 1996 showed that the primary visual cortex of blind people activates during Braille reading. The territory is not available, it is occupied.

The literature is not univocal for all that, and this must be said. Work from the Project Prakash programme in India shows, in patients operated on late after eight to seventeen years of blindness, unexpected gains in contrast sensitivity, independent of age at surgery: substantial plasticity persists. But the same work also shows that after late restoration, static figural cues remain largely ineffective for segmenting an image, and that visual integration relies first on motion, with an improvement spread over ten to eighteen months.

One major reservation must accompany all of this: none of these studies concerns a cortical prosthesis. They involve repaired eyes — most often an operated congenital cataract — feeding an intact visual pathway. Transposing their conclusions to Blindsight is an extrapolation, and must be presented as such.

What to remember

  • Blindsight is a serious research project, part of a real and long-standing scientific lineage. It is not a sham.
  • Nor is it a treatment. As of 17 August 2026, no human data has been published, no vision trial is registered, no authorisation has been issued.
  • What a cortical implant produces, at best, is a coarse perception of shapes and contrasts based on phosphenes. The word “see” does not have here the meaning spontaneously attached to it.
  • It is intended for people who have lost the use of both eyes and of the optic nerves — a situation that does not correspond to that of patients followed for AMD, diabetic retinopathy, glaucoma or cataract.
  • The question to ask today is not “when will the chip arrive?”, but “are my current follow-up and treatment up to date?”.

Unsure about your retinal follow-up or about a change in your vision?

Book an appointment for an assessment

Cachan practice · Tel. +33 1 45 47 08 11
Paris 13 practice · Tel. +33 1 89 31 30 60

Frequently asked questions

Can Neuralink really restore sight to the blind?

Not as things stand. No clinical data from a Blindsight implant in humans has been published in a peer-reviewed journal, and no Neuralink vision trial is registered on the public clinicaltrials.gov registry. What has been demonstrated, by other teams, is the production of elementary light perceptions allowing some letters and the outlines of objects to be identified — in one single patient, for six months, in 2021.

What exactly does a person fitted with a cortical implant see?

Phosphenes: points of light with no stable colour and no sharp outline, whose position in the visual field depends on that of the electrodes. An implant does not deliver an image but a pattern of dots, which the brain has to learn to interpret. Phosphenes also move with eye movements, which remains an unresolved problem.

What is the difference with the PRIMA retinal implant?

It is fundamental. PRIMA is a subretinal implant: it replaces the photoreceptors but uses the patient’s inner retina and optic nerve, which must be functional. It is intended for patients with atrophic AMD, who are not blind and retain peripheral vision. Blindsight is a cortical implant: it bypasses the eye and the optic nerve entirely, and targets people in whom these no longer transmit anything. The two devices are not intended for the same patients.

Is Blindsight approved by the FDA?

No. In September 2024 Neuralink announced that it had obtained the FDA designation of Breakthrough Device. This status speeds up exchanges with the agency and prioritises the review of a future submission, but US law states explicitly that it alters no evaluation criterion. According to FDA figures as of 31 March 2026, 1,284 designations have been granted for 198 marketing authorisations.

Can this treat my AMD or my diabetic retinopathy?

No, and it is not a question of timing. A cortical implant is designed for people whose eyes and optic nerves no longer transmit any signal. In AMD or diabetic retinopathy, the visual pathway remains partly functional: it is the existing retinal treatments — intravitreal injections, laser, surgery — that are indicated, and their benefit depends on the moment at which they are carried out.

Should I wait for this technology before being treated?

That course of action exposes you to permanent visual loss. Destroyed retina does not regenerate, and no implant, available or announced, restores vision comparable to what a treatment carried out in time makes it possible to preserve. A delayed retinal treatment does not recover what has been lost in the meantime.

Does the Argus II bionic eye still exist?

The device has not been produced since 2019 and its manufacturer, Second Sight, has ceased its activity in this field. According to an investigation published by IEEE Spectrum in 2022, more than 350 people worldwide carry an implant that has become obsolete and unsupported. It is a precedent that every patient should know about before considering an implant offered by a private company.

When will this type of implant be available in France?

No date can seriously be put forward. No visual cortical implant holds a marketing authorisation, in Europe or in the United States, and the most advanced academic devices are at the early feasibility stage, in a handful of patients. The question of reimbursement therefore does not arise at this stage.

Sources

  1. Brindley GS, Lewin WS. The sensations produced by electrical stimulation of the visual cortex. The Journal of Physiology. 1968;196(2):479-493. PMID: 4871047
  2. Fernández E, Alfaro A, Soto-Sánchez C, et al. Visual percepts evoked with an intracortical 96-channel microelectrode array inserted in human occipital cortex. The Journal of Clinical Investigation. 2021;131(23):e151331. PMID: 34665780
  3. Beauchamp MS, Oswalt D, Sun P, et al. Dynamic stimulation of visual cortex produces form vision in sighted and blind humans. Cell. 2020;181(4):774-783.e5. PMID: 32413298
  4. Chen X, Wang F, Fernández E, Roelfsema PR. Shape perception via a high-channel-count neuroprosthesis in monkey visual cortex. Science. 2020;370(6521):1191-1196. PMID: 33273097
  5. Fine I, Boynton GM. A virtual patient simulation modeling the neural and perceptual effects of human visual cortical stimulation, from pulse trains to percepts. Scientific Reports. 2024;14(1):17400. PMID: 39075065
  6. Cha K, Horch K, Normann RA. Simulation of a phosphene-based visual field: visual acuity in a pixelized vision system. Annals of Biomedical Engineering. 1992;20(4):439-449. PMID: 1510295
  7. Sommerhalder J, Rappaz B, de Haller R, et al. Simulation of artificial vision: II. Eccentric reading of full-page text and the learning of this task. Vision Research. 2004;44(14):1693-1706. PMID: 15136004
  8. da Cruz L, Dorn JD, Humayun MS, et al. Five-year safety and performance results from the Argus II retinal prosthesis system clinical trial. Ophthalmology. 2016;123(10):2248-2254. PMID: 27453256
  9. Humayun MS, Dorn JD, da Cruz L, et al. Interim results from the international trial of Second Sight’s visual prosthesis. Ophthalmology. 2012;119(4):779-788. PMID: 22244176
  10. Holz FG, Le Mer Y, Muqit MMK, et al. Subretinal photovoltaic implant to restore vision in geographic atrophy due to AMD. The New England Journal of Medicine. 2026;394(3):232-242. PMID: 41124203
  11. Sadato N, Pascual-Leone A, Grafman J, et al. Activation of the primary visual cortex by Braille reading in blind subjects. Nature. 1996;380(6574):526-528. PMID: 8606771
  12. Kalia A, Lesmes LA, Dorr M, et al. Development of pattern vision following early and extended blindness. Proceedings of the National Academy of Sciences. 2014;111(5):2035-2039. PMID: 24449865
  13. Ostrovsky Y, Meyers E, Ganesh S, et al. Visual parsing after recovery from blindness. Psychological Science. 2009;20(12):1484-1491. PMID: 19891751
  14. Okun MS, Marjenin T, Ekanayake J, et al. Definition of implanted neurological device abandonment: a systematic review and consensus statement. JAMA Network Open. 2024;7(4):e248654. PMID: 38687486
  15. Strickland E, Harris M. Their bionic eyes are now obsolete and unsupported. IEEE Spectrum. 15 February 2022. Available at: spectrum.ieee.org/bionic-eye-obsolete
  16. Food and Drug Administration. Breakthrough Devices Program. Legal framework: 21 U.S.C. §360e-3 (Federal Food, Drug, and Cosmetic Act §515B). Available at: govinfo.gov
  17. ClinicalTrials.gov. Trials registered by Neuralink Corporation (NCT06429735, NCT06700304, NCT07127172, NCT06992596, NCT07224256, NCT06710626). Accessed 17 August 2026.
  18. Neuralink Corporation. Neuralink receives Breakthrough Device Designation for Blindsight. 17 September 2024. Available at: neuralink.com
  19. Neuralink Corporation. Two Years of Telepathy. 28 January 2026. Available at: neuralink.com
  20. Uyeno G. Neuralink’s Blindsight implant won’t deliver natural sight. IEEE Spectrum. 27 September 2024. Available at: spectrum.ieee.org
  21. Newsweek. Elon Musk’s remarks on Blindsight made at a public meeting in Wisconsin. 31 March 2025.

Further reading

Disclaimer

This article is intended for information purposes. A personalised ophthalmological opinion remains essential for any treatment decision.

No cortical visual implant is available in France, either in routine care or within a clinical trial open to recruitment in 2026. Dr Tourabaly fits neither cortical implants nor retinal prostheses, and this article in no way constitutes an offer of a procedure. The results of the studies cited hold for the populations and the precise conditions described in each publication, and are not generalisable to other clinical situations. Consultations take place at the Cachan practice and at the Paris 13 practice; surgical procedures are performed in clinic.

Related articles

Written and reviewed by Dr Moïse Tourabaly, ophthalmic refractive surgeon — former chef de clinique (Quinze-Vingts National Eye Hospital).

Last updated: August 24, 2026

Add this site to your Google preferred sources

Similar Posts