The cursor moves, a letter appears, a game of chess begins. In demonstrations of brain–machine interfaces, everything seems to hinge on that moment: an intention becomes an action, without any hand movement. But for a person with paralysis, the real test comes the next day. Will the system still work without a technician nearby? Who will respond if the controls become imprecise? Innovation does not stop in the operating room: it has to last.
Looking ahead to September 2026, this question offers a more useful lens than the race for performance alone. The factual examples discussed here draw on publicly known results and events through 2024; the developments envisaged for 2026 are forward-looking analysis, not an updated clinical assessment.
After the breakthrough, the test of everyday life
In 2024, Neuralink publicly shared the first uses of its implant by Noland Arbaugh, a person with tetraplegia. Controlling a computer, playing games, browsing: these activities put a human face on a technology long confined to laboratories. Synchron is exploring another route, with electrodes delivered through blood vessels and positioned near motor regions, rather than inserted directly into brain tissue.
These approaches are not interchangeable. They differ in their implantation methods, the signals they capture and their hardware constraints. They nevertheless share an ambition: to turn brain activity into a useful command. And they face the same obstacle: a one-off success guarantees neither independent use nor lasting benefit.
Before these demonstrations, work by the BrainGate consortium and other academic teams had already shown possibilities for computer control and communication. Research published in 2023 also illustrated progress in decoding attempted speech in people with paralysis. But performance achieved with a handful of participants under supervised conditions does not yet amount to a service available to everyone.
The brain is not a USB port
An implanted interface does not receive a perfectly stable digital stream. It measures biological phenomena, and software then learns to associate them with intentions. Yet that relationship changes over time. Depending on the device, tiny electrode movements, tissue responses or changes in neural activity can alter the quality of the recorded information.
Fatigue, attention, posture and the surrounding environment can also affect use. A cursor that is easy to control during a short session will not necessarily remain so after several hours. Decoding must accommodate this variability without turning every day into a technical training session.
Neuralink provided a public illustration of this in spring 2024: the company reported that some threads in its first human implant had retracted, reducing the number of effective electrodes. It described software adjustments to improve performance. This episode shows that software can sometimes partially compensate for a hardware problem without eliminating its physical cause.
Recalibrating without exhausting the user
Calibration means retraining the system to interpret signals. It may require imagining a movement, attempting an action or following visual instructions. Repeated too often, it becomes a burden. Progress will therefore also be measured by how long it takes before someone can simply write a message.
Adaptive decoders could reduce this burden. But their autonomy raises a question: how can they distinguish a genuine change in the signal from a temporary misinterpretation? Software that adjusts itself must remain predictable. For the user, regaining control, undoing a command or returning to a reliable setting matters just as much as gaining speed.
Maintenance becomes a medical issue
The implant is surrounded by an entire chain of components: power supply, data transmission, receiver, computer, application and accessories. A faulty cable or an incompatible update can disable a system whose brain-facing component still works. Reliability must be judged across the whole system, not just the electrodes.
This reality calls for a different approach to maintenance than that used for an ordinary connected device. A failure can remove an essential means of communication. Diagnosis must distinguish between a software problem, an external component failure and a complication requiring medical advice. Technical support must therefore be coordinated with clinical follow-up.
In a scenario of wider adoption, procedures for replacing external components, accessible support and secure updates would need to be put in place. Remote access could speed up troubleshooting, but would have to be strictly controlled. Brain data and the commands inferred from it are sensitive, even though these interfaces cannot freely read thoughts.
The patient is not the device technician
At home, a successful interface must fit in with care, getting around, rest and social relationships. If a family member has to spend considerable time setting up the equipment every morning, resolving disconnections and calling the laboratory, some of the promised independence is simply shifted into invisible work.
Support may involve doctors, occupational therapists, alternative communication specialists and technical teams. Priority uses must be chosen with the person: having conversations, working, managing their environment or returning to a leisure activity. The best settings are not necessarily those that produce the most spectacular demonstration.
It is also important to retain fallback options: eye-tracking control where this remains possible, an adapted switch or a communication board. A brain interface should expand the options, not make its user dependent on a single channel. This complementary approach also provides a safeguard on days when fatigue or a technical problem prevents its use.
Who provides ongoing support when the trial ends?
Clinical trials provide intensive support that is difficult to replicate at scale. Any eventual transition to more widespread use will therefore depend on more than regulatory approval alone. Follow-up care will need funding, professionals will need training, and responsibilities will have to be coordinated among manufacturers, healthcare institutions and service providers.
Continuity after the trial is a particularly sensitive issue. What happens if the company changes strategy, goes out of business or abandons a generation of hardware? Removing an implant may require surgery and is not always the preferred solution. Commitments regarding support, data and medical options should be discussed before implantation.
To assess these technologies, a few indicators would be more revealing than a speed record: time spent using the system genuinely independently, recalibration frequency, downtime, caregiver burden and perceived benefit. Publishing these measures over long periods would help compare devices without confusing experimental achievement with everyday usefulness.
What comes next? Looking ahead to 2026 and beyond, differentiation could shift from electrode density to the quality of the service delivered. Those able to stabilize decoding, simplify maintenance and guarantee lasting support would have a decisive advantage. The most convincing demonstration would then be less spectacular: a person using their interface at home, then moving on to something else, without having to think about the technology helping them.
