Choosing music without help. Writing to a loved one. Saying that something hurts when speech is no longer possible. Before promises of enhanced intelligence, brain–machine interfaces pursue a more fundamental goal: restoring agency to people unable to move or communicate. For this feature set in September 2026, we draw on publicly available findings established through mid-2024; the developments discussed are forward-looking projections, not a verified account of approvals or commercial availability in 2026. This is an essential precaution in a field where videos of remarkable achievements often run ahead of clinical evidence.
A cursor on a screen, independence regained
The principle seems simple: capture brain activity, identify a signal associated with an intention, then translate it into a command. The system can move a cursor, select a letter or control a prosthetic limb. It does not read thoughts at will. Within a defined setting, it learns to recognize patterns associated with a task, requiring calibration and delivering variable performance.
For someone with tetraplegia, directly controlling a computer can transform daily life. But brain interfaces are not entering an empty field: eye-tracking controls, switches and sip-and-puff devices already provide considerable help. The right comparison is not with having no assistance, but with the best solution suited to the patient. An implant must offer enough benefit to justify its risks and demands.
Real breakthroughs on different fronts
From imagined movement to decoded speech
Research by the BrainGate consortium has long shown that people with paralysis can use signals recorded in the cortex to control digital or robotic tools. In 2023, two studies published in Nature also marked a milestone in communication research: implanted devices decoded attempted speech in female participants with severe paralysis, producing text or a synthetic voice accompanied by an avatar.
These findings are important, but their scope must be clearly defined. They involve a very small number of people in specialized protocols. Speed, errors, available vocabulary and training requirements vary between systems. A sentence correctly decoded during a session does not yet guarantee spontaneous, reliable conversation available throughout the day at home.
Walking again: a bridge rather than a repair
Another milestone came in 2023, when a team involving EPFL and CHUV, among others, presented a “digital bridge” in Nature linking the brain and spinal cord of a man with a spinal cord injury. Brain signals guided spinal cord stimulation, allowing him to walk again with assistance. This was neither a cure for the injury nor a method already suitable for widespread use.
The result illustrates an essential distinction: some interfaces replace a control pathway; others form part of rehabilitation that may promote functional recovery. These benefits are not interchangeable. Walking with technological assistance, recovering movement without assistance and gaining independence at home are three different measures.
Implant or stay on the surface?
Implanted electrodes can capture more localized signals than those placed on the scalp. But “implant” covers several approaches: electrodes that penetrate the cortex, systems placed on its surface, or devices introduced into a blood vessel near the brain. Each involves trade-offs between signal quality, access to target areas, risks and durability.
Neuralink brought extensive media attention to the field when its first human participant received an implant in 2024. The company later reported that some of the device’s threads had retracted, affecting signal collection before software adjustments were made. The episode is a reminder that the surgical achievement is only the beginning. Synchron, meanwhile, was exploring an endovascular approach: a different access strategy, not an absence of risk or equivalent performance.
Non-invasive interfaces, particularly those based on electroencephalography, avoid surgery. They can be used to select commands, communicate or support certain rehabilitation protocols. In return, their signals are more sensitive to movement, muscle activity and setup conditions. A headset that photographs well may still be difficult to use alone: positioning, contact quality and fatigue matter just as much as the algorithm.
What really needs to be demonstrated
Approval to conduct a trial is not approval to bring a device to market. Nor does a scientific publication mean that treatment costs will be covered. Assessing an announcement requires looking at participant numbers, follow-up duration, adverse events and conditions of use. Peak performance tells us less than the consistency of the assistance provided.
- Relevance: can the patient accomplish a task that matters to them better than with their usual assistive tools?
- Reliability: does the system remain usable despite fatigue, errors and signal variations?
- Safety: what are the surgical, infection-related and hardware risks, and what would removal entail?
- Actual independence: how much human assistance is needed to set up, calibrate and maintain the equipment?
A gain can be valuable even if it seems modest. Being able to reliably answer yes or no may matter more than spectacular typing speeds. Conversely, a device that performs well in the laboratory may lose its value if it requires lengthy preparation every morning. Trials must therefore incorporate quality of life, caregiver burden and user preferences, not just technical scores.
The device does not end with the implant
Access also depends on infrastructure: a surgical team, rehabilitation, maintenance, support and funding. Someone living far from a specialist center does not automatically benefit from an innovation described in a publication. For invasive systems, continuity of follow-up is crucial. Who will provide updates if the company disappears? Who will handle a malfunction or implant removal?
Brain data also raise questions about consent and privacy. Without imagining that the mind can be read in full, combining these data with medical and behavioral information warrants strong protection. Patients must understand what is recorded, transmitted and reused. They must also be able to clearly distinguish participation in research from a promise of therapeutic benefit.
Human enhancement can wait
Applications intended for people without disabilities receive considerable media attention, but their benefit–risk balance is entirely different. Accepting an intervention to regain a means of communication is not the same as accepting one to operate an app faster. Prioritizing medical needs does not reflect a lack of ambition: it requires judging innovation by the difference it actually makes in someone’s life.
What next? The most credible path forward would involve larger trials, longer follow-up and documented use at home. Simpler devices could broaden access; better decoders could reduce the learning effort required. None of this should be presented as a given. The real turning point will not simply be a connected brain in a demonstration, but a patient able to rely on their interface every day, with sustained support.
